Report Look-alike humans identified by facial recognition algorithms show genetic similarities Graphical abstract Authors Ricky S. Joshi, Maria Rigau, Carlos A. Garcı́a-Prieto, ..., Xavier Binefa, Alfonso Valencia, Manel Esteller Correspondence mesteller@carrerasresearch.org In brief We recognize each other by relying on our face uniqueness. However, there are humans with uncanny resemblance. Joshi et al. reported that look-alike pairs identified by facial recognition algorithms share genotypes but not DNA methylomes and microbiomes. The identified SNPs also provide a readout of other anthropomorphic and behavioral characteristics. Highlights d Facial recognition algorithms identify ‘‘look-alike’’ humans for multiomics studies d Intrapair look-alikes share common genetic sequences such as face trait variants d DNA methylation and microbiome profiles only contribute modestly to human likeness d The identified SNPs impact physical and behavioral phenotypes beyond facial features Joshi et al., 2022, Cell Reports 40, 111257 August 23, 2022 ª 2022 The Author(s). https://doi.org/10.1016/j.celrep.2022.111257 ll ll OPEN ACCESS Report Look-alike humans identified by facial recognition algorithms show genetic similarities Ricky S. Joshi,1,10 Maria Rigau,2,10 Carlos A. Garcı́a-Prieto,1,2,10 Manuel Castro de Moura,1 David Piñeyro,1,3 Sebastian Moran,1 Veronica Davalos,1 Pablo Carrión,4 Manuel Ferrando-Bernal,4 Iñigo Olalde,4 Carles Lalueza-Fox,4 Arcadi Navarro,4,5,6 Carles Fernández-Tena,7 Decky Aspandi,8 Federico M. Sukno,8 Xavier Binefa,8 Alfonso Valencia,2,6 and Manel Esteller1,3,6,9,11,* 1Josep Carreras Leukaemia Research Institute (IJC), Badalona, 08916 Barcelona, Spain 2Barcelona Supercomputing Center (BSC), 08034 Barcelona, Spain 3Centro de Investigacion Biomedica en Red Cancer (CIBERONC), 28029 Madrid, Spain 4Institute of Evolutionary Biology (CSIC-Universitat Pompeu Fabra), 08003 Barcelona, Spain 5Centre for Genomic Regulation (CNAG-CRG), 08003 Barcelona, Catalonia, Spain 6Institucio Catalana de Recerca i Estudis Avançats (ICREA), 08010 Barcelona, Spain 7Herta Security, S.L., 08037 Barcelona, Spain 8Departament de Tecnologies de la Informació i les Comunicaciones (DTIC), Universitat Pompeu Fabra (UPF), 08018 Barcelona, Spain 9Physiological Sciences Department, School of Medicine and Health Sciences, University of Barcelona (UB), L’Hospitalet, 08907 Barcelona, Spain 10These authors contributed equally 11Lead contact *Correspondence: mesteller@carrerasresearch.org https://doi.org/10.1016/j.celrep.2022.111257 SUMMARY The human face is one of the most visible features of our unique identity as individuals. Interestingly, mono- zygotic twins share almost identical facial traits and the same DNA sequence but could exhibit differences in other biometrical parameters. The expansion of the world wide web and the possibility to exchange pictures of humans across the planet has increased the number of people identified online as virtual twins or doubles that are not family related. Herein, we have characterized in detail a set of ‘‘look-alike’’ humans, defined by facial recognition algorithms, for their multiomics landscape. We report that these individuals share similar genotypes and differ in their DNA methylation and microbiome landscape. These results not only provide in- sights about the genetics that determine our face but also might have implications for the establishment of other human anthropometric properties and even personality characteristics. INTRODUCTION gard, one of the most documented cases is the Dutch famine at the end of WWII that was associated with less DNA methyl- The discussion about the relevance of ‘‘nature versus nurture,’’ ation of the imprinted IGF2 gene compared with their unex- or, in a similar manner, of ‘‘genotype versus phenotype,’’ in hu- posed, same-sex siblings (Heijmans et al., 2008). man biology and medicine is a long-standing issue that still re- Human individual identity also relates to biological properties mains largely unsolved. Relevant studies in this area include and environment. In this regard, the way we initially recognize our original observation that monozygotic twins show epigenetic each other relies often on our unique face, and there is a sophis- differences (Fraga et al., 2005), understood as the chemical ticated brain code to distinguish facial identities (Tsao et al., marks such as DNA methylation and histone modifications that 2006; Chang and Tsao, 2017; Quian Quiroga, 2017). This ex- regulate gene expression, that might explain different population plains why so commonly twins catch our attention and are traits and distinct penetrance of diseases in these people, a used to understand how the balance between nature and nurture finding supported in later studies (Kaminsky et al., 2009), generates a phenotype. Here, we present a study that, on a mo- including The NASA Twins Study (Garrett-Bakelman et al., lecular level, aims to characterize random human beings that 2019). These questions can be more easily addressed in exper- objectively share facial features. This extraordinary set of individ- imental models where the researcher can intervene, such as the uals, characterized by their high likeliness, are what are called, in Agouti mice (Wolff et al., 1998) and cloned animals (Rideout lay-language, look-alike humans, unknown twins, twin et al., 2001), whereas in humans, the investigator has a more strangers, doubles, or doppelgänger, in German. This unique passive role, waiting for the right sample to appear. In this re- set of samples has allowed us to study how genomics, Cell Reports 40, 111257, August 23, 2022 ª 2022 The Author(s). 1 This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). ll OPEN ACCESS Report Figure 1. Recruitment and objective determination of look-alike human pairs (A) Representation of the global worldwide distribution of 32 look-alike pairs (n = 64) in this study. (B) 27 facial parameters by which the Microsoft Oxford Project face API (Microsoft) objectively performs face detection. (C) Venn diagram showing the number of look-alike pairs discerned and jointly identified in the three facial recognition programs: MatConvNet, Custom-Net, and Microsoft. Numbers within the semi-circle present the pairs that did not cluster in each software. (D) Boxplots showing unbiased quantitative similarity scores comparing each facial recognition software (MatConvNet, Custom-Net, Microsoft) for monozygotic twins (MZs; blue), look-alike pairs (LALs; rose), and random non-LALs (red). The x axis represents the different cohorts analyzed. The y axis exhibits similarity scores measured between 0 and 1. 1 represents identical facial image, and 0 represents two totally different photographic entities. ‘‘N’’ indicates the number of couples. Differences calculated using two-sided Mann-Whitney-Wilcoxon test: ****p < 0.0001; ***p < 0.001; ns, non-significant. (E) Photographic examples of LALs used in this study. epigenomics, and microbiomics can contribute to human resem- ticipants completed a comprehensive biometric and lifestyle blance. Our study provides a rare insight into human likeness by questionnaire in their native language (English, Spanish, and showing that people with extreme look-alike faces share com- French) (Methods S1). Their geographic locations are shown in mon genotypes, whereas they are discordant at their epigenome Figure 1A. We first determined an objective measure of ‘‘like- and microbiome. Genomics clusters them together, and the rest ness’’ for the candidate double pairs. We used three different set them apart. These findings do not only provide clues about methods of facial recognition: the custom deep convolutional the genetic setting associated with our facial aspect, and prob- neural network Custom-Net, (www.hertasecurity.com), the ably other traits of our body and personality, but also highlight MatConvNet algorithm (Vedaldi and Lenc 2015), and the Micro- how much of what we are, and what defines us, is really inherited soft Oxford Project face API (https://azure.microsoft.com/es-es/ or instead is acquired during our lifetime. services/cognitive-services/face/) (STAR Methods). We used three methods because each system can yield variable results, RESULTS and we selected those systems to reflect the diversity of possible outcomes. MatConvNet was designed for facial classification, Facial recognition algorithms and multiomics Custom-Net for surveillance, and Microsoft API for generalized approaches for look-alike humans facial analysis. These models have millions of learned parame- Human doubles were recruited from the photographic work of ters and have been trained with millions of facial images from François Brunelle, a Canadian artist who has been obtaining thousands of subjects, in a variety of unconstrained situations: worldwide pictures of look-alikes since 1999 (http://www. differences in pose, hairstyle, expression, age, and accessories francoisbrunelle.com/webn/e-project.html). We obtained head- within a subject. Thus, the impact of these attributes is likely min- shot pictures of thirty-two candidate look-alike couples. All par- imal. Each software provides a facial similarity score between 2 Cell Reports 40, 111257, August 23, 2022 ll Report OPEN ACCESS Figure 2. Genetic analysis of look-alike human pairs (A) Saliva DNA was obtained from 32 LALs recruited to this study. DNA was subjected to genotyping (Omni5-4 SNP arrays Illumina), DNA methylation (Infinium MethylationEPIC arrays, Illumina), and microbiome analysis (16S Metagenomics sequencing, Illumina). (B) Heatmap of hierarchical genetic clustering with bootstrap of genome-wide SNP genotyping arrays in the 16 LALs. Genotype clustering was performed using Euclidean distances and Ward.D2 cluster method. Blue rectangles represent 9 LALs that unbiasedly clustered. 0 = homozygous reference SNPs (green), 1 = heterozygous SNPs (black), and 2 = homozygous alternate SNPs (red). (C) Boxplot showing Kinship scores between MZs, LALs, and random non-LALs. Kinship scores range between 0.2 (it represents two unrelated individuals) and 0.5 (it represents duplicated genotypes and MZs). ‘‘N’’ indicates the number of couples. Differences calculated using two-sided Student’s t test: ****p < 0.0001; **p < 0.01. (D) Gene Ontology (GO) analysis performed using all SNPs found to be shared in all LALs (19,277 SNPs in 3,730 genes). GO enrichments were ran using EnrichGO R package for the 3,730 genes, and the top 10 most significant hits are plotted in network graphs. GO terms are presented with circles. The size and color of each circle represents numbers of genes in each GO term and its statistical significance, respectively. The gray lines represent the interaction of genes, and the thick- ness is proportional to the number of genes interacting in each GO term. GO subcategories are presented: Biological Process, Cellular Component, and Molec- ular Function. 0 and 1, where 1 is the same facial image and 0 is two different cantly higher than those observed in random non-look-alike entities. Comparisons are pairwise, with every image compared pairs (Figure 1D). Thus, these highly look-alike humans were with every other image. As an example of the parameters the focus of our further research. Illustrative examples of these computed, the 27 face landmarks of the Microsoft algorithm ‘‘double’’ individuals are shown in Figure 1E. are shown in Figure 1B. The results obtained from the different Saliva DNA for these cases was analyzed by multiomics at combinations of each approach are shown in a Venn diagram three levels of biological information: genome, by means of an in Figure 1C. Interestingly, the number of pairs that were consid- SNP microarray that interrogates 4,327,108 genetic variants ered to be correlated by at least two of the facial models was very selected from the International HapMap and 1,000 Genomes high (25 out of total 32, >75%), closer to the human ability to Projects, which target genetic variation down to 1% minor allele recognize identical twins (Biswas et al., 2011). Most importantly, frequency (MAF) (Xing et al., 2016); epigenome, using a DNA we found that 16 of the original 32 (50%) look-alike pairs were methylation microarray that studies over 850,000 CpG sites matched by all three facial recognition systems. As an internal (Moran et al., 2016); and microbiome, by ribosomal RNA direct positive control for high similarity score, we ran the three facial sequencing (Klindworth et al., 2013) (Figure 2A; STAR Methods). recognition software in monozygotic twin photograph images from the University of Notre Dame twins database 2009/2010 Genomic characterization of look-alike humans (https://cvrl.nd.edu/projects/data/). Importantly, similarity scor Genomic analyses of these 16 couples provided a striking result: es from the 16 look-alike couples were similar to those obtained more than half (9 of 16, 56.2%) of these look-alike pairs clustered from monozygotic twins according to MatConvNet and signifi- together in the unsupervised clustering heatmap with bootstrap Cell Reports 40, 111257, August 23, 2022 3 ll OPEN ACCESS Report (Figure 2B). These nine couples were denominated as ‘‘ultra’’ pair of ultra-look-alikes shared the genotype. For example, look-alike. K-means algorithm represented by principal-compo- where one individual in a pair was heterozygous for a given nent analysis (PCA) and t-distributed stochastic neighbor SNP, the corresponding individual in the pair was also heterozy- embedding (t-SNE) also showed that the look-alike couples gous. This genotype match must be consistent across all pairs that clustered by the unsupervised clustering heatmap analysis for an SNP to be considered shared and therefore represented were in close proximity (Figure S1), indicating a likely genotyping indicative SNPs relevant for look-alike resemblance. The number resemblance of the studied pairs. In contrast, the 16 candidate of shared SNP positions was significantly higher compared with look-alike cases that did not cluster by the three facial recogni- random non-look-alike pairs in the studied population tion (FR) networks (Figure 1C) showed that only one pair clus- (p < 2.2 3 10 16, Pearson’s chi-squared test). Taking into ac- tered together (1 of 16, 6.2%) (Figure S1). count ethnicity, shared SNP positions by the European ultra- We studied two possible confounding factors: population look-alike pairs was significantly higher compared with random stratification (ancestry) and kinship. Using KING Relationship non-look-alike pairs in the studied population (p = 0.03, Pear- Inference (Manichaikul et al., 2010) to determine kinship scores, son’s chi-squared test). For the remaining three ethnicities, we discarded the possibility of unknown familial relationships only one individual from each group was available in our dataset. (first and second degree) between look-alike pairs (Figure 2C). Thus, we interrogated the individuals genotyped in the 1000 Ge- We observed that look-alike pairs were more similar to non- nomes database (https://www.internationalgenome.org/). The look-alike pairs than to monozygotic twins (Figure 2C); support- number of shared SNP positions by the Hispanic ultra-look-alike ing that look-alike pairing in the SNP clustering is not related to pair was significantly higher compared with random individual familyhood genotype but instead to a distinct subset of genetic pairs from the same ethnicity (p < 2.2 3 10 16, Pearson’s chi- similarity. Using PLINK (Purcell et al., 2007) (STAR Methods), squared test). No significant enrichment was observed for the re- close kinship could be excluded in almost all cases: only one maining two couples, one East Asian and one Central-South pair share SNPs in proportions that could be compatible with Asian. Importantly, only 16 variants of the 19,277 SNPs up to third-degree relatives and only one pair share a long (>10 (0.08%) selected from the ultra-look-alikes presented a linkage cM) identity by descent (IBD) segment that could suggest co- disequilibrium detected by iterative pruning analysis (Weir ancestry in the last few hundreds of years. Interestingly, the latter et al., 2014). is a French-Canadian pair, a population known to have experi- The identified genetic variants might have a profound impact enced a dramatic founder effect in the 17th century. Importantly, on the degree of similitude between the phenotype of humans. when we conducted all the downstream analyses without this Using the clusterProfiler R package (Yu et al., 2012), we per- French-Canadian pair, the remaining eight ultra-look-alike pairs formed gene enrichment analyses using the list of look-alike clustered together (Figure S2). The detailed kinship assessment SNPs compared with the background of all genes annotated in data are provided in Table S1. the SNP microarray. We observed an enrichment for Gene Related to population stratification, among the 16 look-alike Ontology (GO) Biological Processes related to anatomical, pairs, 13 were of European ancestry, 1 Hispanic, 1 East Asian, developmental, and adhesion terms (Figure 2D; Table S3), in and 1 Central-South Asian. Although background genetic addition to ion and anion binding for GO-Molecular function ancestry is a principal determinant for genetic variance between (gene subsets related to bone and skin properties) and many human populations, we observed that of the 13 White look-alike cellular compartments. Enrichment analysis using the DAVID pairs, 7 (54%) did not cluster genetically, suggesting alternative signature database collection noted that the most significantly purposes for shared genetic variation between look-alike pairs. enhanced ontology was ‘‘cell junction,’’ a critical determinant To further determine ancestry, genotyping of the 16 look-alike of tissue morphology (Table S4). To evaluate the face genes cases was performed using GenomeStudio v.2.0.5 to create enrichment in our selected 19,277 SNPs corresponding to PACKPED Plink files (STAR Methods). Their genomic data 3,730 genes (Table S2), we gather all the genes related with were merged with 1,980 West Eurasian, Asian, and Native Amer- face traits from recent data (Claes et al., 2018; Xiong et al., ican individuals genotyped in the Affymetrix Human Origins (HO) 2019; White et al., 2021), Facebase dataset (https://www. array (Lazaridis et al., 2014), where the remaining dataset held facebase.org/), and Genome-wide Association Study (GWAS) 175,469 common SNPs. PCA was generated with the HO indi- Central (study HGVST1841, http://www.gwascentral.org) and viduals (Figure S3) and look-alike individuals (Figure S3B for applied a hypergeometric test and a Monte Carlo simulation us- West Eurasia and Figure S3C for West Eurasia, Asia, and Amer- ing 10,000 iterations (STAR Methods). In no iteration of random ica) (Price et al., 2006; Patterson et al., 2006) (STAR Methods). set of genes did we observe a number equal to or higher than We observed that almost all the look-alike pairs cluster close the face genes represented in our 19,277 SNP selection to each other according to their countries of origin (or self-attrib- (p < 1e 4). We observed a total of 1,794 face genes in our uted ethnic background) (Figure S3). However, they are not more 19,277 SNP selection, constituting 26% of all the face genes pre- closely related than other pairs of individuals from the same pop- sent in the array (hypergeometric test p: 6.31e 172; Monte Carlo ulations taken at random. The detailed population stratification empirical p < 1e 4). When we added the reported face associ- data are provided in Table S1. ated SNPs to our 19,277 SNPs, we observed that 11 of the 16 Among the 9 couples of ultra-look-alikes, 19,277 SNP posi- (68.7%) look-alike pairs clustered together (Figure S4), therefore tions annotated for 3,730 genes (Table S2) were defined as adding two new couples. SNPs with shared genotypes in each look-alike pair. These The study of the functional nature of the SNPs loci shared by SNPs correspond to non-monomorphic positions in which every the ultra-look-alikes showed that 171 caused amino acid 4 Cell Reports 40, 111257, August 23, 2022 ll Report OPEN ACCESS changes, affecting 158 genes (Table S5). GOrilla analysis for GO- genotyping (Figure 2B). The comparison of DNA methylation pat- Molecular function found an enrichment in anion transport terns among the nine look-alike couples with the observed descriptors (Table S3). Using the GWAS catalog database genetic overlap (Figure 2B) only clustered one additional pair (https://www.ebi.ac.uk/gwas/), we found that 113 SNPs corre- (Figure S4). K-means algorithm represented using PCA and the sponded to 130 GWAS associations and 84 traits (Table S6). t-SNE plot did not show significant clustering (Figure S5). These last traits included many related to facial determinants Thus, overall, human look-alikes are diverse in their epigenome or physical features such as cleft palate/lip, eye color, hip settings. circumference, body height, waist-hip ratio, balding measure- However, two avenues might provide a role for DNA methyl- ment, and alopecia (Table S6) with an enrichment for lip and fore- ation in facial morphology: epigenetic age and methylation QTL head morphology, body mass index, bone mineral density, and (meQTLs). The aging process changes facial morphology, and attached earlobe (Table S6). We observed an enrichment of traits DNA methylation is used as a proxy for ‘‘biological age’’ that that included the word morphologytagged to the terms nose, lip, can or can not be directly related to the ‘‘chronological age.’’ mouth, facial, cranial vault, forehead, hair, and cheekbone One example is the premature epigenetic aging observed in car- (Fisher’s exact test, odds ratio [OR] = 4.2, p = 0.04). Using the riers of viral infections (Esteban-Cantos et al., 2021; Cao et al., GWAS Central database (http://www.gwascentral.org), we 2022). We have calculated the intrapair absolute age differences found an enrichment (OR = 1.2782, p = 0.0007364) for SNPs in our 16 look-alike cohort according to chronological age (date associated with human facial variation (Adhikari et al., 2016). of birth) or epigenetic age (DNA methylation clock) (Hannum The analyses of the look-alike SNPs according to trait in et al., 2013). We found no differences in intrapair chronological GWAS Central showed an enrichment for the phenotype names age between the ultra-look-alike group and the non-ultra-look- ‘‘lip’’ (OR = 1.8321, p = 0.000327) and ‘‘forehead’’ (OR = 1.886, alike group. In contrast, intrapair ‘‘epigenetic’’ age differences p = 0.010389). The identified look-alike SNPs were also enriched were smaller among ultra-look-alike pairs compared with the (OR = 2.201156, p = 0.04884) for genes included in the FaceBase non-ultra-look-alike group (two-sided Mann-Whitney-Wilcoxon dataset (https://www.facebase.org/). Finally, we studied the test, p = 0.0052) (Figure S6). DNA methylation is also associated overlap between the herein discovered look-alike SNPs and with genetic variation (Villicaña and Bell, 2021) and could expression quantitative trait loci (eQTLs). Using the Genotype- contribute to individual similarity acting as meQTLs. Using the Tissue Expression (GTEx) Portal (https://www.gtexportal.org/ methylation status of 1,379 CpG sites located within a window home/), we observed that look-alike SNPs were more frequently of +100 bp from the identified 19,277 SNPs, we observed that associated with gene-expression changes than expected by 3 of the 16 (18.7%) look-alike pairs clustered together (Figure S6). random chance (Fisher’s exact test, OR = 1.1, p = 0.0001). The All three of these pairs were among the 9 ultra-look-alike couples enrichment was observed among different morphological struc- (Figure 2B). Thus, DNA methylation, as a marker of biological age tures and organs (Table S6). We also used the stratified linkage and meQTL, can also provide phenotypic commonality for ultra- disequilibrium score regression (S-LDSC) (Finucane et al., look-alikes. 2015) to determine the enrichment of GWAS signals from the A similar scenario was found for the microbiome. From a qual- GWAS catalog for our SNPs. We observed that these SNPs itative standpoint (alpha diversity), according to the type of bac- were overrepresented for the pronasale-right chelion (enrich- teria present in the studied oral sample (STAR Methods), only ment score [ES] = 13.84, p = 0.018) and pronasale-left chelion one look-alike pair clustered together (Figure 3C). This couple (ES = 12.26, p = 0.04) face traits (Figure S4) (Xiong et al., did not cluster together according to SNP genotyping (Figure 2B). 2019). The SNPs were also overrepresented for features that From a quantitative standpoint, according to the amount of each define 63 facial segments (Hoskens et al., 2021) considering bacteria strand present (STAR Methods), we found clustering of the entire, mid, and outer face (p < 0.05) (Figure S4). These one look-alike pair (6.25%, 1 of 16) (Figure 3D). This couple also data indicate that the 19,277 characterized SNPs exert a major paired together by unsupervised SNP clustering (Figure 2B). The impact in the way the face of humans is defined. study of the nine couples with SNP similarity did not provide The SNP microarray can also be used to determine copy-num- further pairing of look-alikes (Figure S6). K-means algorithm illus- ber variations (CNVs) (Feber et al., 2014). Unsupervised clus- trated by PCA and t-SNE did not demonstrate clustering (Fig- tering heatmap with bootstrap clustered only one couple ure S7). Thus, look-alikes do not mostly share a microbiome. together of the 16 look-alikes according to CNVs (Figure 3A). However, oral microbiome relates to obesity (Yang et al., Interestingly, three CNVs were shared by three look-alike pairs 2019), and fat in the face could relate to similarities. We found (Table S6), including a locus in chromosome 11 that targets that intrapair weight differences were smaller among ultra- genes involved in craniofacial dysmorphic features such as look-alike pairs compared with non-ultra-look-alike pairs (two- HYLS1 (Mee et al., 2005). sided Student’s t test test, p = 0.035) (Figure S7). Thus, it is possible that the oral microbiome, through its relation to fat con- Other multiomics views of look-alike humans tent, contributes to look-alike phenotypes. Similar ‘‘identities’’ of look-alikes could also reside in other ‘‘omic’’ components such as the DNA methylome and the micro- Traits of look-alike humans beyond facial features biome. According to DNA methylation patterns, only one of the The likeness between the identified human pairs is not limited to sixteen (6.25%) look-alike pairs matched both individuals the shared facial traits. All the recruited participants in the study together, as shown in the unsupervised clustering heatmap (Fig- completed a comprehensive biometric and lifestyle questionnaire ure 3B). This couple also clustered together according to SNP (Methods S1), and the collected information is summarized in Cell Reports 40, 111257, August 23, 2022 5 ll OPEN ACCESS Report Figure 3. Copy-number variation, DNA methylation, and microbiome analysis of LALs (A) Heatmap shows the hierarchical clustering of the samples based on the copy number (scale of 0–4) of all copy-number variation (CNV) regions, defined as regions in which at least one individual carried a different copy number. A random selection of one-fifth of such CNV regions is represented in this plot, but the clustering of samples had been obtained considering all CNV regions. The blue rectangle represents a LAL that clusters together. (B) Heatmap shows unsupervised genome-wide DNA methylation hierarchical clustering with bootstrap of the 16 LALs, using the methylation b-values obtained from MethylationEPIC arrays. A random selection of 5000 CpGs is represented. Colors represent a continuous quantification of methylation beta values at each CpG site, where green highlights unmethylated CpGs (0), black, 50% methylated CpGs (0.5), and red, fully methylated CpGs (1). Clustered look-alikes are shown in a blue rectangle. (C and D) Microbiome analysis of 16 LALs. Heatmaps show the distances from differences in pairwise bacterial counts of species found in the microbiome of each LAL (variation in alpha diversity scores) of counts from 0–55 (3C) and relative proportions of the taxonomic profiles at the genus level (3D) for each sample calculated on a scale of 0–0.5. Only the most represented genera are shown. Meta-genomic clustering of each look-alike sample was constructed using Euclidean distances and Ward.D2 hierarchical cluster method. Blue rectangle represents LALs whose microbiome is closely related. Figure 4A. Overall, 68 parameters (Table S7) were included and (p = 0.00006) (Figure 4B). Examples of independent questionnaire converted to numerical or logical (0/1) variables (STAR Methods, variables (such as height, weight, smoking habit, or level of educa- (custom scripts GitHub: https://github.com/mesteller-bioinfolab/ tion) further demonstrate that look-alike pairs are closer than non- lookalike). The input curated questionnaire is shown in Table S7. look-alike pairs (Figure 4C). Thus, humans with a similar face might We used a cosine similarity method (STAR Methods) to calculate also share a more comprehensive physical, and probably behav- likeness between the studied individuals according to the ques- ioral, phenotype that relates to their shared genetic variants. Our tionnaire answers. Studying the original 32 look-alike couples, study supports the concept of heritability estimation that individ- we observed that the 16 look-alike pairs that matched together uals correlated at the phenotype level share a significant number by all three facial recognition software showed shorter Euclidean of genotypic correlations (Visscher et al., 2008). Our results are distances within pairs (p = 0.03475) and higher cosine similarity germane to the ongoing efforts to predict biometric traits scores (p = 0.00321) than those pairs that did not match by the from genomic data (Lippert et al., 2017) and the diagnosis of ge- facial algorithms (Figure 4B). According to their SNPs, the 16 netic disorders using facial analysis technologies (Gripp et al., look-alike pairs showed shorter Euclidean distances compared 2016; Hadj-Rabia et al., 2017; Hsieh et al., 2019; Gurovich et al., with those pairs that did not match by the three facial algorithms 2019). 6 Cell Reports 40, 111257, August 23, 2022 ll Report OPEN ACCESS Figure 4. Biometric and lifestyle analysis of LALs using cosine similarity scores (A) Representation of the biometric and lifestyle parameters considered to calculate cosine similarity scores. (B) Euclidean distances between the individuals from a pair (intra-pair distance) compared with the distance between individuals from different pairs (extra-pair distance). Distances were calculated for questionnaire (top) and SNP data (below). Statistics by Student’s t test. (C) Distance boxplots for independent questionnaire variables generated by calculating, for all possible pairs of samples, their absolute differences for each variable. We then classified all pairs between pairs of look-alikes and pairs of non-look-alikes. Statistics by Wilcoxon rank sum tests. DISCUSSION Non-genetic factors can affect the expression of genes that form the face. Many epigenetic or imprinting disorders present Our study deciphers molecular components associated with craniofacial anomalies, such as patients with Prader-Willi or An- facial construction by applying a multiomics approach in a gelman syndrome (Girardot et al., 2013), and microbial disrup- unique cohort of look-alike humans that are genetically unre- tion is associated with developmental defects (Robertson lated. Saliva DNA was subjected to genome-wide analyses of et al., 2019). Despite evidence for epigenetic variation in human common genetic variation, DNA methylation, and microbiome populations (Heyn et al., 2013) and development (Garg et al., analysis. We also performed a biometric and lifestyle analysis 2018), only one look-alike pair clustered by DNA methylation. for all look-alike pairs. We found that 16 of the 32 look-alike pairs This pair also clustered together by SNPs, suggesting that the clustered in all three facial recognition software. Genetic analysis shared epigenetic profile is likely due to their underlining shared revealed that 9 of these 16 look-alike pairs (Figure 2B) clustered, genetics (Lienert et al., 2011), as it was also supported by identifying 19,277 common SNPs. Furthermore, analyses of analyzing CpGs in the vicinity of the SNPs. In addition, ultra- these shared variants in GWAS and GTEx databases revealed look-alike pairs showed similar epigeneticages. Similarly, only enrichment for phenotypes related to body and face structures one look-alike pair clustered by microbiome analysis, but ultra- and an association with gene-expression changes. Together, look-alike pairs displayed similar weights, and microbiome this suggests that shared genetic variation in humans that look composition could relate to obesity (Yang et al., 2019). These alike likely contribute to the common phenotype. findings support a modest role for these biological components Historically, research into face morphology was heavily to determine facial shape; however, more evidence is required to centered on craniofacial anomalies (Richmond et al., 2018). discard a greater impact. However, there is a recent growing interest into normal-range Finally, 68 biometric and lifestyle attributes from the look-alike face variation, attributable to the necessity for facial recognition pairs were studied. Physical traits such as weight and height as software for everyday life (smartphones, CCTV cameras, etc.). well as behavioral traits such as smoking and education were Easy access to low-cost, high-resolution pictures and advances correlated in look-alike pairs, suggesting that shared genetic in genotyping technology has ignited an age-old question: what variation not only relates to shared physical appearance but makes humans look as they do? Association studies revealed may also influence common habits and behavior. low-frequency genetic variants with relatively small penetrance Overall, we provided a unique insight into the molecular char- in facial features, suggesting a far more complex genetic role. acteristics that potentially influence the construction of the Cell Reports 40, 111257, August 23, 2022 7 ll OPEN ACCESS Report human face. We suggest that these same determinants correlate B Multiomics clustering analyses with both physical and behavioral attributes that constitute hu- B Questionnaires processing and similarity analysis man beings. These findings provide a molecular basis for future applications in various fields such as biomedicine, evolution, and SUPPLEMENTAL INFORMATION forensics. Through collaborative efforts, the ultimate challenge would be to predict the human face structure based on the indi- Supplemental information can be found online at https://doi.org/10.1016/j. celrep.2022.111257. vidual’s multiomics landscape. ACKNOWLEDGMENTS Limitations of the study Due to the difficulty to obtain look-alike data and biomaterial, the We thank François Brunelle for providing the look-alike images. We thank sample size is small, restricting our ability to perform large-scale CERCA Programme/Generalitat de Catalunya and the Josep Carreras Foun- statistical analyses. Thus, some partially negative results, such dation for institutional support. This work was funded by the governments of as those derived from the non-genetic data, could relate to an Catalonia (2017SGR1080) and Spain (RTI2018-094049-B-I00, SAF2014- 55000, and TIN2017-90124-P) and the Cellex Foundation. underpowered study. The used headshots were two-dimen- sional, black and white images, and valuable information AUTHOR CONTRIBUTIONS regarding three-dimensional constructs, subtle skin tones, and unique facial features are lacking. In addition, the used SNP array M.E. conceived and designed the study; R.S.J., M.R., C.A.G.-P., M.C.d.M., does not allow for the analysis of other genetic components such D.P., S.M., V.D., P.C., M.F.-B., I.O., C.L.-F., A.N., C.F.-T., D.A., F.M.S., X.B., as structural variations and shared rare events. Another limita- A.V., and M.E. analyzed multiomics and questionnaire data; R.J. and M.E. tion is that our samples were mostly from European origin. wrote the manuscript with contributions and approval from all authors. Thus, the study could not effectively address the impact of the used multiomics in other human populations. DECLARATION OF INTERESTS M.E. is a consultant of Ferrer International and Quimatryx. S.M. is an employee STAR+METHODS of Ferrer International. C.F.-T. is chief technical officer of Herta Security. Detailed methods are provided in the online version of this paper Received: July 16, 2021 and include the following: Revised: June 5, 2022 Accepted: August 1, 2022 d KEY RESOURCES TABLE Published: August 23, 2022 d RESOURCE AVAILABILITY B Lead contact REFERENCES B Materials availability Adhikari, K., Fuentes-Guajardo, M., Quinto-Sánchez, M., Mendoza-Revilla, J., B Data and code availability Camilo Chacón-Duque, J., Acuña-Alonzo, V., Jaramillo, C., Arias, W., Lozano, d EXPERIMENTAL MODEL AND SUBJECT DETAILS R.B., Pérez, G.M., et al. (2016). A genome-wide association scan implicates B Recruitment of look-alikes DCHS2, RUNX2, GLI3, PAX1 and EDAR in human facial variation. Nat. Com- d METHOD DETAILS mun. 7, 11616. https://doi.org/10.1038/ncomms11616. B Facial recognition algorithms Aryee, M.J., Jaffe, A.E., Corrada-Bravo, H., Ladd-Acosta, C., Feinberg, A.P., B Facial similarity Hansen, K.D., and Irizarry, R.A. (2014). Minfi: a flexible and comprehensive B Sample preparation Bioconductor package for the analysis of Infinium DNA methylation microar- rays. Bioinformatics 30, 1363–1369. https://doi.org/10.1093/bioinformatics/ B HumanOmni5-Quad BeadChip btu049. B Infinium MethylationEPIC BeadChip Beck, T., Shorter, T., and Brookes, A.J. (2020). GWAS Central: a comprehen- B 16S meta-genomics sequencing sive resource for the discovery and comparison of genotype and phenotype d QUANTIFICATION AND STATISTICAL ANALYSIS data from genome-wide association studies. Nucleic Acids Res. 48, D933– B Population-level vs shared SNPs in look-alike pairs D940. https://doi.org/10.1093/nar/gkz895. B Copy number variant (CNV) calling and functional Biswas, S., Bowyer, K.W., and Flynn, P.J. (2011). A study of face recognition of annotation identical twins by humans. In IEEE International Workshop on Information Fo- B CNV clustering and heatmap rensics and Security, Iguacu Falls, Brazil, pp. 1–6. https://doi.org/10.1109/ B Genome-wide SNP arrays from monozygotic twins WIFS.2011.6123126. B Cryptic relatedness Buniello, A., MacArthur, J.A.L., Cerezo, M., Harris, L.W., Hayhurst, J., Malan- B Ancestry assessment gone, C., McMahon, A., Morales, J., Mountjoy, E., Sollis, E., et al. (2019). The NHGRI-EBI GWAS catalog of published genome-wide association studies, B Kinship assessment targeted arrays and summary statistics. Nucleic Acids Res. 47, D1005– B Functional enrichment of shared SNPs using Gene D1012. https://doi.org/10.1093/nar/gky1120. Ontology Cao, X., Li, W., Wang, T., Ran, D., Davalos, V., Planas-Serra, L., Pujol, A., Es- B Face gene enrichment in the identified SNPs teller, M., Wang, X., and Yu, H. (2022). Accelerated biological aging in COVID- B GWAS analysis 19 patients. Nat. Commun. 13, 2135. https://doi.org/10.1038/s41467-022- B GWAS functional enrichment of shared SNPs using S- 29801-8. LDSC Chang, L., and Tsao, D.Y. (2017). The code for facial identity in the primate B DNA methylation age estimation brain. Cell 169, 1013–1028.e14. https://doi.org/10.1016/j.cell.2017.05.011. 8 Cell Reports 40, 111257, August 23, 2022 ll Report OPEN ACCESS Claes, P., Roosenboom, J., White, J.D., Swigut, T., Sero, D., Li, J., Lee, M.K., Heyn, H., Moran, S., Hernando-Herraez, I., Sayols, S., Gomez, A., Sandoval, Zaidi, A., Mattern, B.C., Liebowitz, C., et al. (2018). Genome-wide mapping of J., Monk, D., Hata, K., Marques-Bonet, T., Wang, L., et al. (2013). DNA methyl- global-to-local genetic effects on human facial shape. Nat. Genet. 50, ation contributes to natural human variation. Genome Res. 23, 1363–1372. 414–423. https://doi.org/10.1038/s41588-018-0057-4. https://doi.org/10.1101/gr.154187.112. Eden, E., Lipson, D., Yogev, S., and Yakhini, Z. (2007). Discovering motifs in Hoskens, H., Liu, D., Naqvi, S., Lee, M.K., Eller, R.J., Indencleef, K., White, ranked lists of DNA sequences. PLoS Comput Biol. 3, e39. J.D., Li, J., Larmuseau, M.H.D., Hens, G., et al. (2021). 3D facial phenotyping by biometric sibling matching used in contemporary genomic methodologies. Eden, E., Navon, R., Steinfeld, I., Lipson, D., and Yakhini, Z. (2009). GOrilla: a PLoS Genet. 17, e1009528. https://doi.org/10.1371/journal.pgen. tool for discovery and visualization of enriched GO terms in ranked gene lists. BMC Bioinformatics. 10, 48. Hsieh, T.C., Mensah, M.A., Pantel, J.T., Aguilar, D., Bar, O., Bayat, A., Becerra- Solano, L., Bentzen, H.B., Biskup, S., Borisov, O., et al. (2019). PEDIA: priori- Esteban-Cantos, A., Rodrı́guez-Centeno, J., Barruz, P., Alejos, B., Saiz-Me- tization of exome data by image analysis. Genet. Med. 21, 2807–2814. https:// drano, G., Nevado, J., Martin, A., Gayá, F., De Miguel, R., Bernardino, J.I., doi.org/10.1038/s41436-019-0566-2. et al. (2021). Epigenetic age acceleration changes 2 years after antiretroviral therapy initiation in adults with HIV: a substudy of the NEAT001/ANRS143 Huang, D.W., Sherman, B.T., and Lempicki, R.A. (2009). Systematic and inte- randomised trial. Lancet. HIV 8, e197–e205. https://doi.org/10.1016/S2352- grative analysis of large gene lists using DAVID bioinformatics resources. Nat. 3018(21)00006-0. Protoc. 4, 44–57. https://doi.org/10.1038/nprot.2008.211. Feber, A., Guilhamon, P., Lechner, M., Fenton, T., Wilson, G.A., Thirlwell, C., Kaminsky, Z.A., Tang, T., Wang, S.C., Ptak, C., Oh, G.H.T., Wong, A.H.C., Morris, T.J., Flanagan, A.M., Teschendorff, A.E., Kelly, J.D., et al. (2014). Using Feldcamp, L.A., Virtanen, C., Halfvarson, J., Tysk, C., et al. (2009). DNA high-density DNA methylation arrays to profile copy number alterations. methylation profiles in monozygotic and dizygotic twins. Nat. Genet. 41, Genome Biol. 15, R30. https://doi.org/10.1186/gb-2014-15-2-r30. 240–245. https://doi.org/10.1038/ng.286. Keegan, K.P., Glass, E.M., and Meyer, F. (2016). MG-RAST, a metagenomics Finucane, H.K., Bulik-Sullivan, B., Gusev, A., Trynka, G., Reshef, Y., Loh, P.R., service for analysis of microbial community structure and function. Methods Anttila, V., Xu, H., Zang, C., Farh, K., et al. (2015). Partitioning heritability by Mol. Biol. 1399, 207–233. https://doi.org/10.1007/978-1-4939-3369-3_13. functional annotation using genome-wide association summary statistics. Nat. Genet. 47, 1228–1235. https://doi.org/10.1038/ng.3404. Klindworth, A., Pruesse, E., Schweer, T., Peplies, J., Quast, C., Horn, M., and Glöckner, F.O. (2013). Evaluation of general 16S ribosomal RNA gene PCR Fortin, J.-P., Triche, T.J., Jr., and Hansen, K.D. (2017). Preprocessing, normal- primers for classical and next-generation sequencing-based diversity studies. ization and integration of the Illumina HumanMethylationEPIC array with minfi. Nucleic Acids Res. 41, e1. https://doi.org/10.1093/nar/gks808. Bioinformatics 33, 558–560. https://doi.org/10.1093/bioinformatics/btw691. Lazaridis, I., Patterson, N., Mittnik, A., Renaud, G., Mallick, S., Kirsanow, K., Fraga, M.F., Ballestar, E., Paz, M.F., Ropero, S., Setien, F., Ballestar, M.L., Heine- Sudmant, P.H., Schraiber, J.G., Castellano, S., Lipson, M., et al. (2014). Suñer, D., Cigudosa, J.C., Urioste, M., Benitez, J., et al. (2005). Epigenetic differ- Ancient human genomes suggest three ancestral populations for present- ences arise during the lifetime of monozygotic twins. Proc. Natl. Acad. Sci. USA day Europeans. Nature 513, 409–413. https://doi.org/10.1038/nature13673. 102, 10604–10609. https://doi.org/10.1073/pnas.0500398102. Lienert, F., Wirbelauer, C., Som, I., Dean, A., Mohn, F., and Schu€beler, D. Garrett-Bakelman, F.E., Darshi, M., Green, S.J., Gur, R.C., Lin, L., Macias, (2011). Identification of genetic elements that autonomously determine DNA B.R., McKenna, M.J., Meydan, C., Mishra, T., Nasrini, J., et al. (2019). The methylation states. Nat. Genet. 43, 1091–1097. https://doi.org/10.1038/ng. NASA Twins Study: a multidimensional analysis of a year-long human space- 946. flight. Science 364, eaau8650. https://doi.org/10.1126/science.aau8650. Lippert, C., Sabatini, R., Maher, M.C., Kang, E.Y., Lee, S., Arikan, O., Harley, Garg, P., Joshi, R.S., Watson, C., and Sharp, A.J. (2018). A survey of inter-in- A., Bernal, A., Garst, P., Lavrenko, V., et al. (2017). Identification of individuals dividual variation in DNA methylation identifies environmentally responsive co- by trait prediction using whole-genome sequencing data. Proc. Natl. Acad. regulated networks of epigenetic variation in the human genome. PLoS Genet. Sci. USA 114, 10166–10171. https://doi.org/10.1073/pnas.1711125114. 14, e1007707. https://doi.org/10.1371/journal.pgen.1007707. Manichaikul, A., Mychaleckyj, J.C., Rich, S.S., Daly, K., Sale, M., and Chen, Girardot, M., Feil, R., and Llères, D. (2013). Epigenetic deregulation of genomic W.M. (2010). Robust relationship inference in genome-wide association imprinting in humans: causal mechanisms and clinical implications. Epige- studies. Bioinformatics 26, 2867–2873. https://doi.org/10.1093/bioinformat- nomics 5, 715–728. https://doi.org/10.2217/epi.13.66. ics/btq559. Gripp, K.W., Baker, L., Telegrafi, A., and Monaghan, K.G. (2016). The role of McDonald, D., Price, M.N., Goodrich, J., Nawrocki, E.P., DeSantis, T.Z., objective facial analysis using FDNA in making diagnoses following whole Probst, A., Andersen, G.L., Knight, R., and Hugenholtz, P. (2012). An improved exome analysis. Report of two patients with mutations in the BAF complex Greengenes taxonomy with explicit ranks for ecological and evolutionary ana- genes. Am. J. Med. Genet. 170, 1754–1762. https://doi.org/10.1002/ajmg.a. lyses of bacteria and archaea. ISME J. 6, 610–618. https://doi.org/10.1038/is- 37672. mej.2011.139. Gurovich, Y., Hanani, Y., Bar, O., Nadav, G., Fleischer, N., Gelbman, D., Basel- Mee, L., Honkala, H., Kopra, O., Vesa, J., Finnilä, S., Visapää, I., Sang, T.K., Salmon, L., Krawitz, P.M., Kamphausen, S.B., Zenker, M., et al. (2019). Iden- Jackson, G.R., Salonen, R., Kestilä, M., and Peltonen, L. (2005). Hydrolethalus tifying facial phenotypes of genetic disorders using deep learning. Nat. Med. syndrome is caused by a missense mutation in a novel gene HYLS1. Hum. Mol. 25, 60–64. https://doi.org/10.1038/s41591-018-0279-0. Genet. 14, 1475–1488. https://doi.org/10.1093/hmg/ddi157. Hadj-Rabia, S., Schneider, H., Navarro, E., Klein, O., Kirby, N., Huttner, K., Moran, S., Arribas, C., and Esteller, M. (2016). Validation of a DNA methylation Wolf, L., Orin, M., Wohlfart, S., Bodemer, C., et al. (2017). Automatic recogni- microarray for 850, 000 CpG sites of the human genome enriched in enhancer tion of the XLHED phenotype from facial images. Am. J. Med. Genet. 173, sequences. Epigenomics 8, 389–399. https://doi.org/10.2217/epi.15.114. 2408–2414. https://doi.org/10.1002/ajmg.a.38343. Mu €llner, D. (2013). Fastcluster: fast hierarchical, agglomerative clustering rou- Hannum, G., Guinney, J., Zhao, L., Zhang, L., Hughes, G., Sadda, S., Klotzle, tines for R and Python. J. Stat. Softw. 53, 1–18. https://doi.org/10.18637/jss. B., Bibikova, M., Fan, J.B., Gao, Y., et al. (2013). Genome-wide methylation v053.i09. profiles reveal quantitative views of human aging rates. Mol. Cell 49, Parkhi, O.M., Vedaldi, A., and Zisserman, A. (2015). Deep face recognition. In 359–367. https://doi.org/10.1016/j.molcel.2012.10.016. Proceedings of the British Machine Vision Conference (BMVA Press), Heijmans, B.T., Tobi, E.W., Stein, A.D., Putter, H., Blauw, G.J., Susser, E.S., pp. 1–12. https://doi.org/10.5244/C.29.41. Slagboom, P.E., and Lumey, L.H. (2008). Persistent epigenetic differences Patterson, N., Price, A.L., and Reich, D. (2006). Population structure and eige- associated with prenatal exposure to famine in humans. Proc. Natl. Acad. nanalysis. PLoS Genet. 2, e190. https://doi.org/10.1371/journal.pgen. Sci. USA 105, 17046–17049. https://doi.org/10.1073/pnas.0806560105. 0020190. Cell Reports 40, 111257, August 23, 2022 9 ll OPEN ACCESS Report Price, A.L., Patterson, N.J., Plenge, R.M., Weinblatt, M.E., Shadick, N.A., and Villicaña, S., and Bell, J.T. (2021). Genetic impacts on DNA methylation: Reich, D. (2006). Principal components analysis corrects for stratification in research findings and future perspectives. Genome Biol. 22, 127. https://doi. genome-wide association studies. Nat. Genet. 38, 904–909. https://doi.org/ org/10.1186/s13059-021-02347-6. 10.1038/ng1847. Visscher, P.M., Hill, W.G., and Wray, N.R. (2008). Heritability in the genomics Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M.A., Bender, D., era–concepts and misconceptions. Nat. Rev. Genet. 9, 255–266. https://doi. Maller, J., Sklar, P., de Bakker, P.I., Daly, M.J., et al. (2007). PLINK: a tool set org/10.1038/nrg2322. for whole-genome association and population-based linkage analyses. Am. J. Weir, B., Cockerham, C., and Feldman, M. (2014). Complete characterization of Hum. Genet. 81, 559–575. https://doi.org/10.1086/519795. disequilibrium at two loci. In Mathematical Evolutionary Theory (Princeton: Quian Quiroga, R. (2017). How do we recognize a face? Cell 169, 975–977. Princeton University Press), pp. 86–110. https://doi.org/10.1515/978140085 https://doi.org/10.1016/j.cell.2017.05.012. 9832-007. White, J.D., Indencleef, K., Naqvi, S., Eller, R.J., Hoskens, H., Roosenboom, J., R Core Team (2019). In R: A Language and Environment for Statistical Lee, M.K., Li, J., Mohammed, J., Richmond, S., et al. (2021). Insights into the Computing Computer Program, version 3.6. 1. genetic architecture of the human face. Nat. Genet. 53, 45–53. https://doi.org/ Ralph, P., and Coop, G. (2013). The geography of recent genetic ancestry 10.1038/s41588-020-00741-7. across Europe. PLoS Biol. 11, e1001555. https://doi.org/10.1371/journal. Wolff, G.L., Kodell, R.L., Moore, S.R., and Cooney, C.A. (1998). Maternal epi- pbio.1001555. genetics and methyl supplements affect agouti gene expression in Avy/a mice. Richmond, S., Howe, L.J., Lewis, S., Stergiakouli, E., and Zhurov, A. (2018). FASEB J. 12, 949–957. https://doi.org/10.1096/fasebj.12.11.949. Facial genetics: a brief overview. Front. Genet. 9, 462. https://doi.org/10. Xing, C., Huang, J., Hsu, Y.H., DeStefano, A.L., Heard-Costa, N.L., Wolf, P.A., 3389/fgene.2018.00462. Seshadri, S., Kiel, D.P., Cupples, L.A., and Dupuis, J. (2016). Evaluation of po- Rideout, W.M., 3rd, Eggan, K., and Jaenisch, R. (2001). Nuclear cloning and wer of the Illumina HumanOmni5M-4v1 BeadChip to detect risk variants for epigenetic reprogramming of the genome. Science 293, 1093–1098. https:// human complex diseases. Eur. J. Hum. Genet. 24, 1029–1034. https://doi. doi.org/10.1126/science.1063206. org/10.1038/ejhg.2015.244. Xiong, Z., Dankova, G., Howe, L.J., Lee, M.K., Hysi, P.G., de Jong, M.A., Zhu, Robertson, R.C., Manges, A.R., Finlay, B.B., and Prendergast, A.J. (2019). The G., Adhikari, K., Li, D., Li, Y., et al. (2019). Novel genetic loci affecting facial human microbiome and child growth - first 1000 Days and beyond. Trends Mi- shape variation in humans. Elife 8, e49898. https://doi.org/10.7554/eLife. crobiol. 27, 131–147. https://doi.org/10.1016/j.tim.2018.09.008. 49898. Suzuki, R., and Shimodaira, H. (2006). Pvclust: an R package for assessing the Yang, Y., Cai, Q., Zheng, W., Steinwandel, M., Blot, W.J., Shu, X.O., and Long, uncertainty in hierarchical clustering. Bioinformatics 22, 1540–1542. https:// J. (2019). Oral microbiome and obesity in a large study of low-income and Af- doi.org/10.1093/bioinformatics/btl117. rican-American populations. J. Oral Microbiol. 11, 1650597. https://doi.org/ Tsao, D.Y., Freiwald, W.A., Tootell, R.B.H., and Livingstone, M.S. (2006). A 10.1080/20002297.2019.1650597. cortical region consisting entirely of face-selective cells. Science 311, Yu, G., Wang, L.G., Han, Y., and He, Q.Y. (2012). clusterProfiler: an R package 670–674. https://doi.org/10.1126/science.1119983. for comparing biological themes among gene clusters. OMICS 16, 284–287. Vedaldi, A., and Lenc, K. (2015). MatConvNet in Proceedings of the 23rd ACM https://doi.org/10.1089/omi.2011.0118. International Conference on Multimedia (MM ’15) (ACM Press), pp. 689–692. Zhang, C., and Zhang, Z. (2010). A survey of recent advances in face detection. https://doi.org/10.1145/2733373.2807412. Microsoft Research. 10 Cell Reports 40, 111257, August 23, 2022 ll Report OPEN ACCESS STAR+METHODS KEY RESOURCES TABLE REAGENT or RESOURCE SOURCE IDENTIFIER Critical commercial assays Oragene DNA tubes DNA Genotex OG-500 Pico Green fluorescence kit Life technologies/thermos P7589 EZ DNA Methylation Kit Zymo Research D5003 Deposited data HumanOmni5-Quad BeadChip This paper GEO: GSE142304 Infinium MethylationEPIC BeadChip This paper GEO: GSE142304 16S metagenomics sequencing This paper BioProject: PRJNA596439 Custom scripts This paper https://github.com/mesteller-bioinfolab/lookalike Look-alike photographs www.francoisbrunelle.com/ https://github.com/mesteller-bioinfolab/lookalike/ webn/e-project.html blob/master/FB_LAL_images.zip Experimental models: Organisms/strains Humans (Homo sapiens) Look-alike individuals upon N/A consent. Software and algorithms R R Core team., 2019 www.r-project.org/ MatConvNet VLFeat http://www.vlfeat.org/matconvnet Microsoft Oxford Project face API Microsoft Azure https://azure.microsoft.com/en-us/services/ cognitive-services/face/ Herta CNN algorithm Herta Security www.hertasecurity.com GenomeStudio (v2.0.4) Illumina https://support.illumina.com/downloads/ genomestudio-2-0.html pvclust Suzuki and Shimodaira, 2006 http://stat.sys.i.kyoto-u.ac.jp/prog/pvclust/ hclust €llner, 2013 Mu https://stat.ethz.ch/R-manual/R-devel/library/ stats/html/hclust.html Kinship-based INference for Manichaikul et al., 2010 http://people.virginia.edu/wc9c/KING/ GWAS (KING v2.2.3) Minfi (v1.32.0) Aryee et al., 2014 https://bioconductor.org/packages/release/ Fortin et al., 2017 bioc/html/minfi.html clusterProfiler Yu et al., 2012 https://guangchuangyu.github.io/2016/01/ go-analysis-using-clusterprofiler/ Database for Annotation, Huang et al., 2009 https://david.ncifcrf.gov/ Visualization and Integrated Discovery (DAVID v6.8) GOrilla http://cbl-gorilla.cs.technion.ac.il/ Eden et al., 2007, 2009 GTEx portal (v7) https://gtexportal.org/ N/A GWAS catalog Buniello et al., 2019 https://www.ebi.ac.uk/gwas/ GWAS central Beck et al., 2020 https://www.gwascentral.org/ MG-RAST Keegan et al., 2016 https://www.mg-rast.org/ Greengenes rRNA database McDonald et al., 2012 https://greengenes.secondgenome.com/ Other François Brunelle website www.francoisbrunelle.com/webn/ N/A e-project.html University of Notre Dame twins https://cvrl.nd.edu/projects/data/ N/A database 2009/2010 Cell Reports 40, 111257, August 23, 2022 e1 ll OPEN ACCESS Report RESOURCE AVAILABILITY Lead contact Further information and requests for reagents and resource may be directed to and will be fulfilled by the lead contact, Dr. Manel Esteller (mesteller@carrerasresearch.org). Materials availability This study did not generate new unique reagents. Data and code availability d SNP and DNA methylation data have been deposited at GEO and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. Microbiome data have been deposited on the BioProject repository and are pub- licly available as of date of publication. Photographs of the look-alike pairs that were matched together for all three different independent facial recognition softwares have been deposited at GitHub and is publicly available as of the date of publication. The DOI is listed in the key resources table. d Original code has been deposited at GitHub and is publicly available as of the date of publication. DOIs are listed in the key resources table. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. EXPERIMENTAL MODEL AND SUBJECT DETAILS Recruitment of look-alikes 32 Look-alike pairs (n = 64 individuals) that were initially recruited and photographed by François Brunelle (http://www. francoisbrunelle.com/webn/e-project.html) were enrolled to this study. All 64 individuals [42 females (65.6%) and 22 males (34.4%) with a median age of 40 years (range from 21 to 78 years), Table S7] were required to complete an extensive biometric and life-style questionnaire (Methods S1: Data collection questionnaire, related to STAR Methods) as well as provide legally signed consent forms approved by our bioethics committee for usage of both their facial images and DNA samples for this study. The study protocol was approved by the Clinical Research Ethics Committee of the Bellvitge University Hospital with the reference number PR348/16. To compliment this study, we were also provided with access to 100 monozygotic twin photos from the University of Notre Dame twins database 2009/2010 (https://cvrl.nd.edu/projects/data/). License agreements for data access were reviewed and signed by legal representatives of all entities involved in this study. 50 monozygotic twin pairs (n = 100) photographs were subsequently downloaded and analysed with the facial recognition algorithms detailed below. METHOD DETAILS Facial recognition algorithms Three facial recognition algorithms were used to objectively analyze look-alike pairs: MatConvNet CNN algorithm, provided by the University of Pompeu i Fabra, Barcelona (Vedaldi and Lenc 2015); Microsoft Oxford Project face API by Microsoft; and the custom deep convolutional neural network Custom-Net (www.hertasecurity.com). The quantitative assessment of pairwise similarity be- tween face photographs was calculated as follows. For the MatConvNet algorithm, the face biometric template from each photo was extracted from each processed face by means of a deep convolutional neural network (CNN) built into MatConvNet software. The resulting templates are represented as integer sparse descriptors of 8,192 values, which effectively encode the identity features of a face image (Vedaldi and Lenc 2015). Final pairwise similarity scores were set on a scale of 0–1 where 1 represents identical faces. The custom deep convolutional neural network Custom-Net was developed by a leader in facial recognition platforms (www. hertasecurity.com). Firstly, a generic face detector optimized for unconstrained video surveillance scenarios was used to obtain the locations of all faces in each image (Zhang and Zhang, 2010). The threshold was adjusted to find all targeted faces in each photo, and a subsequent manual exploration was conducted to ensure that no false positives were included. Each face was cropped with a 25% extra margin from the original bounding box, converted to grayscale and resized to 250 3 250 pixels. Next, a face biometric template was extracted from each processed face by means of a deep convolutional neural network of 32 layers. The resulting tem- plates were represented as integer sparse descriptors of 4,096 values, which effectively encode the identity features of a face image. Finally, the similarity score between a pair of images was computed as a negative mean square deviation between their template values. The final scores were mapped to a range 0–1, where 1 indicated identical faces, according to landmarks taken from the his- togram of imposter pairs extracted from the well-known database (http://vis-www.cs.umass.edu/lfw/). In the case of the custom deep convolutional neural network, the models have tens of millions of learned parameters and have been trained with more than 10 million facial images from over a hundred thousand subjects from different human populations, in a variety of unconstrained situations: differences of pose, expression, age and accessories within a subject. Moreover, the training process of a face recognition algorithm typically involves "data augmentation" operations, in which input images are randomly modified, e.g. by e2 Cell Reports 40, 111257, August 23, 2022 ll Report OPEN ACCESS artificially synthesizing glasses, adding facial occlusions, mirroring faces, etc. in order to add intraclass variability to the images and confer robustness to the resulting model. As a consequence, modern face verification algorithms have recently achieved near-per- fect accuracy, as high as 99.97% on NIST’s Facial Recognition Vendor Test (https://pages.nist.gov/frvt/html/frvt11.html#overview), for passport photo or mugshot scenarios, to the point that banks worldwide have widely adopted such systems for user verification. Particularly, these algorithms have become extremely reliable on controllable, almost ideal scenarios such as those captured by the photographer: 1:1 verification between large resolution images with good illumination, non-lateral poses (less than 60 ) and without heavy occlusions; despite circumstancial similaritiy in interclass appearance like that given by glasses, facial expression or hairstyle. Thus, the impacts of these attributes, such as pose, hairstyle etc can be considered minimum, because the incorporated models have been exposed to these variations, in addition to additional features aspects such as colour styles, image degradations etc. The VGG dataset (https://www.robots.ox.ac.uk/vgg/data/vgg_face/) shows examples of facial data used to train Matconvnet (Par- khi et al., 2015) and CustomNet (http://vis-www.cs.umass.edu/lfw/). The Microsoft Oxford Project face API by Microsoft operates on a number of attributes that affect facial features such as age, gender, pose, smile, and facial hair along with 27 other landmarks for each face. These landmarks are left pupil, right pupil, nose tip, left mouth, right mouth, outer left eyebrow, inner left eyebrow, outer left eye, top left eye, bottom left eye, inner left eye, inner right eye, outer right eyebrow, inner right eye, top right eye, bottom right eye, outer right eye, left nose root, right nose root, top left nose alar, top right nose alar, left outer tip of nose alar, right outer tip of nose alar, top upper lip, bottom upper lip, top under lip and bottom under lip (https://azure.microsoft.com/en-us/services/cognitive-services/face/). The final similarity scores were also set on a scale of 0–1. Facial similarity Pair-wise facial similarity matrices were provided as an output for all three facial recognition software. Similarity scores were assigned as numerical values ranging between 0 – 1 where 1 represents identical images and 0, two opposed images. To obtain objective look- alike pairs, we performed unsupervised hierarchical clustering with bootstrap using the pvclust (Suzuki and Shimodaira 2006) in R statistical environment (v3.6.1) (https://www.R-project.org/). Sample preparation Genomic DNA from look-alike pairs in this study were isolated from saliva and self-collected into Oragene 500 DNA tubes and ex- tracted according to the manufacturers instructions (DNA genotek). >10% of the extracted DNA corresponded to microbial DNA. DNA was quantified using Pico Green fluorescence kit/Qubit 2.0 Fluorometer (life technologies). Bisulfite modification of genomic DNA was carried out with the EZ DNA Methylation Kit (Zymo Research) following the manufacturer’s protocol. HumanOmni5-Quad BeadChip Comprehensive cross-examination of genome-wide single nucleotide variation of 4.3 million SNVs across all Look-alike pairs was performed using HumanOmni5-Quad BeadChip (Illumina). 400 ng of genomic DNA was applied to HumanOmni5-Quad BeadChip and scanned using HiScan SQ system (Illumina). The signal raw intensities for each array were assessed and analyzed with GenomeStudio Software (v2.0.4) (Illumina) using default normalization to generate X and Y intensity values for A and B alleles (generic labels for two alternative SNP alleles), respectively. Genotype calling were performed by using GenomeStudio GenCall method and only genotypes with high GenCall scores (GC) were selected (according to Illumina standards). The positions corresponding to Illu- mina internal controls were also removed from the analysis. In order to remove the positions shared between look-alike pairs by chance, a bootstrap look-alike control analysis was performed. Briefly, we generated 100 datasets of 16 random pairs extracted from the initial 32 pairs (64 individuals) used in the study and the complete SNP set from the Omni5 array (4M SNPs). The only require- ment was that none of the generated random pairs in the 100 datasets included a candidate look-alike pair from the initial 32 couples. We applied to each of these new 100 "non-look-alike" datasets the same SNP selection protocol used in the look-alike datasets, i.e. removing monomorphic and non-autosomal positions and selecting the shared inter-look-alike genotypes for each of the 16 pairs. This iterative process produced 100 independent SNP datasets that represented shared genotypes between non-look-alike pairs. Each of the SNP lists obtained contained an average of 5000 SNPs. The plot of the cumulative distribution of these shared SNPs after 100 iterations shows that the number of observed SNPs tends to plateau, indicating that we are reaching a maximum number of SNPs shared by the non-look-alike pairs is being reached. Next, we pooled all 100 SNP datasets into one table removing all redundant variants. This table of unique SNPs was considered as the SNP positions shared between pairs independent of their look-alike status (by chance) and were subsequently removed from our analysis of the look-alike pairs. Then the XY and monoallelic positions for the 16 original pairs were removed. Finally, the SNPs with identical genotypes in each of the 16 pairs and located in genes were selected for further analysis. CNV calling was performed by using PennCNV plugin in GenomeStudio with default parameters. Infinium MethylationEPIC BeadChip Genome-wide DNA methylation interrogation of >850,000 CpG sites was performed using the Infinium MethylationEPIC BeadChip (Illumina) according to manufacturer’s recommended protocol, as previously described (Moran et al., 2016). Briefly, 600 ng of DNA was used to hybridize to the EPIC BeadChip and scanned using HiScan SQ system (Illumina). Raw signal intensity data were initially QC’d and pre-processed from resulting idat files in R statistical environment (v3.6.1) using minfi Bioconductor package (v1.32.0). Cell Reports 40, 111257, August 23, 2022 e3 ll OPEN ACCESS Report A number of quality control steps were applied to minimize errors and remove erratic probe signals. Firstly, interrogation of sex chro- mosomes was performed to identify potential labeling errors. Next, the removal of problematic probes was carried out, such as failed probes (detection p value > 0.01), cross-reacting probes and probes that overlapped single nucleotide variants within +/ 1bp of CpG sites followed by background correction and dye-based normalization using ssNoob algorithm (single-sample normal-expo- nential out-of-band). Lastly, we removed all sex chromosomes. Final DNA methylation scores for each CpG were represented as a b-values ranging between standard 0 and 1 where 1 represents fully methylated CpGs and 0, fully unmethylated. All downstream analyses were performed under R statistical environment (v3.6.1). 16S meta-genomics sequencing We identified and compared bacterial populations from diverse microbiomes from all look-alike pairs using 16S metagenomics sequencing (Illumina) (Klindworth et al., 2013). Salival DNA was extracted and bacterial libraries prepared following the Illumina 16S Library preparation protocol. The variable V3 and V4 regions of 16S rRNA was amplified in order to obtain a single amplicon of approximately 460 bp that underwent paired-end sequencing using MiSeqDx (Illumina). Resulting fastq files were analysed using MG-RAST. The counts corresponding to taxonomic abundance profiles for each sample were retrieved by using MG-RAST tools. Particularly, we retrieved the bacterial counts from sequences aligned to Genus taxonomic categories in the Greengenes rRNA data- base with the following cutoffs: an alignment length of 15 bp, a percent identity of 60% and an e-value equal or lower to 1 3 10 5. The relative proportions for each genus and sample were calculated and only the most represented genus were used. QUANTIFICATION AND STATISTICAL ANALYSIS Population-level vs shared SNPs in look-alike pairs In order to define the number of SNPs shared between non look-alike pairs by chance we generated 55 random combinations of the 9 ultra look-alike pairs avoiding in each dataset the presence of a look-alike pair. We selected the SNP positions with the same geno- type for each of the 9 non look-alike pairs in any of the 55 control datasets, obtaining the percent of randomly shared variants in a data set of 9 non look-alikes. Finally, we calculated the statistical significance of the comparison between SNPs shared in look-alike and non look-alike pairs by a Pearson’s chi-squared test (p value <2.2 10 16). However, since different pairs of look-alikes were from mul- tiple different ethnicities, but individuals in the same look-alike pair shared the same ethnicity, we also performed the enrichment analysis to determine if the number of shared SNPs was more than expected by chance accounting to ethnicity. Thus, we tested pairs of European ancestry individuals with other Europeans and repeated the same for each of the different ethnicities. To this end, we downloaded the most recent set of Omni genotypes from 1000 Genomes available in the phase 3 release directory (ftp://ftp. 1000genomes.ebi.ac.uk/vol1/ftp/release/20130502/supporting/hd_genotype_chip/). The downloaded 1000 Genomes phase 3 vcf file was transformed to Genomic Data Structure (GDS) format using the function seqVCF2GDS from SeqArray R package (version 1.36.0). Look-alike PLINK PED files were also transformed to GDS format using the fucntion snpgdsPED2GDS from SNPRelate R package (version 1.30.1). The 1000 Genomes genotyping data was merged with the ‘‘ultra’’ look-alikes genotyping data and the re- maining dataset held 67,312 common SNPs. Finally, for each ethnicity we generated 55 random combinations of non look-alike pairs to test if the number of shared SNPs in our ‘‘ultra’’ look-alike population was more than expected by chance. Considering the Euro- pean ancestry of the majority of ‘‘ultra’’ look-alike (6 out of 9) and non-‘‘ultra’’ look-alike (7 out of 7) pairs in our study, we used the 7 non-‘‘ultra’’ look-alike pairs with European ancestry to create 55 random combinations of 6 random non look-alike pairs to compute the number of shared SNPs with the same genotype as a proxy for the European population. For East Asia, Central-South Asia and Hispanic populations, we generated 55 random combinations of 1 random non look-alike 1000 Genomes pair to compute the number of shared SNPs in each of the aforementioned populations. Finally, the number of SNPs shared by ‘‘ultra’’ look-alike pairs in each population was tested for statistical significance enrichment against the background number of shared SNPs in each non look-alike population by means of the Pearson’s chi-squared test. Copy number variant (CNV) calling and functional annotation The impact of CNVs on genes was calculated in two different ways. First, we looked at whole-gene CNVs, and then partially-over- lapping CNVs. Copy number of all genes in the genome was calculated by first establishing CNV breakpoints. Breakpoints were as- signed to the outermost SNP positions of regions with the same copy number. The breakpoints were calculated separately for each sample. Using these coordinates, the copy number of whole protein-coding and RNA genes was calculated for all individuals. Gene coordinates were obtained from Ensembl v75 (build GRCh37). We took the genes that had a shared copy number in all pairs of look- alikes (both individuals within the pair had the same number of copies), and we selected those genes for which at least one pair of look-alikes had a different number of copies than the rest of the pairs. For example, to look for partially-overlapping CNVs, we selected all positions in the genome in which the copy number matched within all pairs, but for which at least 2 pairs of lookalikes had a different copy number to the rest of the pairs. We then looked for overlaps with partial overlaps with coding or non-coding genes. As an example, region chr11:125778219-125780253, which overlaps with a lncRNA that has a regulatory relationship with the HYLS1 gene, there are three pairs of look-alikes that carry three copies of this lncRNA, while the remaining pairs have two copies of it. All custom R scripts for CNV analysis are deposited in GitHub repository: https://github.com/mesteller-bioinfolab/lookalike. e4 Cell Reports 40, 111257, August 23, 2022 ll Report OPEN ACCESS CNV clustering and heatmap Clustering of CNVs was done after filtering out all positions with the same copy number in all samples and merging all contiguous positions with the same copy number. Positions from the X and Y chromosomes that showed the same copy number in all males and the same copy number in all females were also filtered out. The clustering of the samples was calculated using pvclust (Suzuki and Shimodaira 2006). Variants represented in the heatmap are a random selection of one fifth of the total number of variants. Genome-wide SNP arrays from monozygotic twins We obtained single nucleotide polymorphism (SNP) data for 38 monozygotic twins from two publicly available studies. Both were downloaded from NCBI Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) under accession No. GSE33598 and GSE9608. The signal raw intensities for each array were assessed and analyzed with GenomeStudio Software (v2.0.4) (Illumina) using default normalization to generate X and Y intensity values for A and B alleles (generic labels for two alternative SNP alleles), respec- tively. All downstream analyses were performed in the R statistical environment (v3.6.1) (https://www.R-project.org/). Cryptic relatedness Robust relatedness inference and genetic correlation estimates between monozygotic twins, look-alike pairs and random non look- alikes were calculated using the software KING (Kinship-based INference for GWAS) (version 2.2.3). Student’s t-test was applied to calculate statistical significance between populations. Ancestry assessment Genotyping was performed using GenomeStudio v2.0.5; PACKPED Plink files were created using the software PLINK Input Report Plug-in v2.1.4 (https://emea.support.illumina.com/downloads/genomestudio-2-0-plugins.html). To analyze the look-alike pairs in the context of world-wide genetic diversity, their genomic data was merged using with 1,980 West-Eurasian, Asian and Native Amer- ican individuals genotyped in the Affimetrix HO array (Lazaridis et al., 2014); the remaining dataset held 175,469 common SNPs. Prin- cipal Component Analysis (PCA) was generated with the HO individuals. Look-Alike individuals were then projected onto the first two components (PC1 and PC2) using options ‘lsqproject: YES’ and ‘shrinkmode: YES’ of smartpca built-in module of EIGENSOFT (v. 7.2.1) (Patterson et al., 2006; Weir et al., 2014) (https://www.hsph.harvard.edu/alkes-price/software/). Kinship assessment Kinship coefficients between look-alike pairs was first estimated with PLINK. PLINK uses a method-of-moments approach where the total proportion of shared SNPs IBD is calculated based on the estimated allele frequency of all SNPs in a dataset assumed to be homogeneous (Purcell et al., 2007). PLINK-indep-pairwise option was used with parameters 50 5 1.5. to generate a pruned subset of genotypes in low linkage disequilibrium of 282,122 SNPs in comparisons with 1000G dataset and 103,256 in comparisons with HO dataset; pairwise relatedness between individuals of each pair was calculated with the –genome–min-0.05 command to detect pairs with levels of IBD sharing compatible with up to a 3rd degree relationship (Manichaikul et al., 2010). Potential relatedness between pairs was subsequently explored by estimating long (>10 cM) IBD blocks that might be indicative of co-ancestry among individuals occurring in the last few hundreds or years (Ralph and Coop, 2013). Functional enrichment of shared SNPs using Gene Ontology Enrichment analysis was done with the enrichGO function from the clusterProfiler R package (Yu et al., 2012), using the org.Hs.eg.db genome annotation. The tested 3,730 genes annotated to the 19,277 SNPs with a matching genotype in all pairs of look-alikes. The background list of genes were all genes annotated to SNPs detected in HumanOmni5-Quad BeadChip analysis. Parameters min- GSSize and maxGSSize from the enrichGO function were set to 1 and 22000, respectively, in order to capture all gene ontologies. Additional enrichment analyses were done using DAVID v6.8 and GOrilla. Enrichment of eQTLs in the look-alike SNPs set was calculated using data from the GTEx portal, release v7 (GTEx_Analy- sis_v7.metasoft.txt.gz). eQTLs with a fixed effect model p-values < 0.05 were selected for the analysis. A Fisher’s test was performed to calculate if the overlap between look-alike SNPs and eQTLs was bigger than expected by chance. The same enrichment analysis was done with each tissue independently, considering the eQTLs with a tissue-specific p-value <0.05. Gene ontology analysis was performed using GOrilla. Face gene enrichment in the identified SNPs In order to statistically evaluate the face genes enrichment in our selected 19,277 SNPs corresponding to 3,730 genes shared by all ‘‘ultra’’ look-alike pairs, we gather all the genes related with face traits (face genes) from recent comprehensive genomic screenings related to facial shape (Claes et al., 2018; Xiong et al., 2019; White et al., 2021), the Facebase dataset (https://www.facebase.org/) and GWAS central (study HGVST1841, http://www.gwascentral.org) and applied two different approaches. In the first approach, we applied a hypergeometric test, as it is implemented in the R ‘‘phyper’’ function, from the package ‘‘stats’’. In the second, we also performed a Monte Carlo simulation using 10,000 iterations. In each iteration, we selected a random set of 3,730 genes (the same number of genes in our 19,277 SNPs) from the total genes represented in the array (23,774 genes) and we counted the number of face genes found in this random selection. All the analyses were performed in R statistical programming language v.4.0.3. Cell Reports 40, 111257, August 23, 2022 e5 ll OPEN ACCESS Report GWAS analysis The overlap between matching sets of SNPs called from look-alike pairs and GWAS SNPs was performed using data from two GWAS databases: GWAS Catalog and GWAS Central. In GWAS Catalog v1.0.2, all GWAS SNPs were retrieved and lifted over from GRCh38 to GRCh37 using the R package liftOver. To calculate trait enrichment, we performed Fisher’s exact tests, computing matching ge- notypes from look-alike pairs against all SNPs detected in the HumanOmni5-Quad BeadChip. For GWAS Central analysis, studies related to facial morphology (HGVST1044, HGVST1625, HGVST1841, HGVST1892, HGVST1933, HGVST2265, HGVST2325, HGVST2359, HGVST2363 and HGVST2597) were selected. Fisher’s exact tests were performed to calculate significant overlaps in the different studies and correction for multiple testing was done with Benjamini and Hochberg’s adjustment method (a = 0.05). All custom R scripts for SNP functional analysis are deposited in GitHub repository: https://github.com/mesteller-bioinfolab/ lookalike. GWAS functional enrichment of shared SNPs using S-LDSC In order to determine the enrichment of GWAS signals for specific annotations we used the stratified LD score regression (S-LDSC) tool (github.com/bulik/ldsc). S-LDSC is a method to estimate heritability enrichment for selected functional annotations. To this end, we followed the partitioned heritability analysis tutorial (github.com/bulik/ldsc/wiki/Patitioned-Heritability) using the last and recom- mended version of the baseline-LD model (version 2.2) with 97 annotations. To asses the heritability enrichment of our 19,277 SNPs, we included a ‘‘look-alike’’ custom functional annotation, defined by the set of 19277 SNPs, on top of the baseline-LD model v2.2. Since S-LDSC is typically applied to large annotations, we included a 500-bp window around the set of 19,277 SNPs to define our custom ‘‘look-alike’’ functional annotation category, following the annotation format of the baseline-LD model v2.2. Considering the European ancestry of the majority of samples in our study, we performed the S-LDSC analysis using European LD scores and allele frequencies from the 1000 Genomes Phase 3 project. Full summary statistics available for ‘‘facial morphology’’ trait in European ancestry individuals were downloaded from GWAS Catalog, corresponding to two studies (Xiong et al., 2019; Hoskens et al., 2021). Finally, partition heritability analysis was performed with default parameters and facial traits with ES >1 and enrichment p value < 0.05 were considered. DNA methylation age estimation Epigenetic age estimation was computed using the Hannum method using the function methyAge from the ENmix R package (version 1.32.0). Multiomics clustering analyses To genetically, epigenetically and metagenomically categorize inherent similarities between all look-alike pairs, shared SNV, CNV, DNA methylation and microbiota profiles, robust correlations and unsupervised hierarchical clustering with bootstrapping were per- formed with R function packages pvclust (Suzuki and Shimodaira 2006). Euclidean distance scores and ward.d2 minimum variance method were applied to attain hierarchical clustering represented as heatmaps using R statistical environment (v3.6.1). K-means clustering was also performed and represented using the first two dimensions of a Principal Component Analysis (PCA). To perform k-means clustering, 16 ‘‘centers’’ (clusters) were indicated. The SNP set was also visualized using t-SNE representation, selecting 2 dimensions and adjusting ‘‘perplexity’’ parameter to 6 and ‘‘max_iter’’ to 5,000. All the analysis were performed in R statistical pro- gramming language v.4.0.3 using the packages ‘‘SNPRelate’’, ‘‘gdsfmt’’,‘‘stats’’, ‘‘Rtsne’’, ‘‘ggfortify’’ and ‘‘ggplot2’’. Questionnaires processing and similarity analysis Data obtained through questionnaires was transformed into a table, which was processed and transformed into numerical format with a custom script (deposited in GitHub; https://github.com/mesteller-bioinfolab/lookalike). In this script, all logical variables were transformed to 0 (False/No) and 1 (True/Yes). When the variables could be ordered (e.g. Never - Sometimes - Often), they were assigned numbers (0–1 - 2 in the example) that were afterwards normalized to 1. For non-sortable variables, the categories were split into logical columns (e.g. Employment category was split into three logical variables - Executive, Salaried and Own business). Finally, empty boxes were filled with the mode for each variable. Cosine similarity was calculated using the numerical matrix between all in- dividuals. The look-alike intra and extra-pair distance analysis were defined and calculated as follows. Intra-pairs were defined as look-alike pairs that clustered in all three facial recognition software (n = 16). The extra-pairs were defined as all other combination pairs of non look-alikes in the initial 16 pairs. For 32 individuals, pairs of same individuals and their look-alike pair counterpart were removed, leaving 30 possible combinations per 16 pair (n = 480). The euclidean distances between each individual and all other sam- ples were calculated using the dist function from the R package pvclust (Suzuki and Shimodaira 2006). Distances were calculated on SNP, CNV, methylome, quantitative and qualitative microbiome and questionnaire data. Intra-pair distances were compared to ex- tra-pair distances using Student’s T test. Distance boxplots for independent variables were generated by calculating, for all possible pairs of samples, their absolute differences for each variable. We then classified all pairs between pairs of look-alikes and pairs of non-look-alikes. Finally, we calculated if the differences were significant with Wilcoxon rank sum tests. e6 Cell Reports 40, 111257, August 23, 2022 Cell Reports, Volume 40 Supplemental information Look-alike humans identified by facial recognition algorithms show genetic similarities Ricky S. Joshi, Maria Rigau, Carlos A. García-Prieto, Manuel Castro de Moura, David Piñeyro, Sebastian Moran, Veronica Davalos, Pablo Carrión, Manuel Ferrando- Bernal, Iñigo Olalde, Carles Lalueza-Fox, Arcadi Navarro, Carles Fernández- Tena, Decky Aspandi, Federico M. Sukno, Xavier Binefa, Alfonso Valencia, and Manel Esteller Supplemental information Figure S1. Clustering and visualization methods. Related to Figure 2. Figure S2. Genetic analyses of look-alike human pairs without the French Canadian pair. Related to Figure 2. Figure S3. Principal Component Analysis (PCA) generated with the genomic data from the 1,980 West- Eurasian, Asian and Native American individuals genotyped in the Affymetrix Human origins (HO) array (Lazaridis et al. 2014). Related to Figure 2. Figure S4. Heatmaps and Stratified LD score regression (S-LDSC) enrichment analysis. Related to Figure 2. Figure S5. Alternative clustering and visualization methods for the DNA methylation data. Related to Figure 3. Figure S6. DNA methylation age estimation, mEQTLs and microbiome analyses. Related to Figure 3. Figure S7. Clustering for metagenomics species data and intrapair weight differences. Related to Figure 3. Methods S1. Data collection questionnaire, related to STAR Methods. A B C 0 2 Figure S1. Clustering and visualization methods. Related to Figure 2. A. k-means clustering of the 16 look-alike couples using the whole array SNP set, excluding non-autosomic and monomorphic SNPs (2,409,249 SNPs). Each color represents a cluster found by the k-means algorithm. B. t-distributed Stochastic Neighbor Embedding (t-SNE) visualization of the same SNP data. Each look-alike couple is depicted in a different color. Red circles were drawn to highlight the position of the 9 ultra look-alike couples in the plot. C. Heatmap of hierarchal genetic clustering with bootstrap of genome-wide SNP genotyping arrays in the 16 look-alikes that were not clustered by the three facial recognition networks. Genotype clustering was performed using Euclidean distances and ward.D2 cluster method. Blue rectangle represents one look-alike pair that unbiasedly clustered. 0 = homozygous reference SNPs (green), 1 = heterozygous SNPs (black) and 2 = homozygous alternate SNPs (red). A 0 2 B C Figure S2. Genetic analyses of look-alike human pairs without the French Canadian pair. Related to Figure 2. A. Heatmap of hierarchical genetic clustering with bootstrap of genome-wide SNP genotyping arrays in the 15 look-alikes. Genotype clustering was performed using Euclidean distances and ward.D2 cluster method. 0 = homozygous reference SNPs (green), 1 = heterozygous SNPs (black) and 2 = homozygous alternate SNPs (red). B. k-means clustering of the 15 look-alike couples using the whole array SNP set, excluding non-autosomic and monomorphic SNPs (2,409,249 SNPs). Each color represents a cluster found by the k-means algorithm. C. t-distributed Stochastic Neighbor Embedding (t-SNE) visualization of the same SNP data. Each look-alike couple is depicted in a different color. Red circles were drawn to highlight the position of the 8 ultra look-alike couples in the plot. A B C PCA Look−Alike (West−Eurasia) PCA Look−Alike −− West Eurasia−Asia−America −Look−Alike Pairs− −Look−Alike Pairs− 0.04 LAL−0003 (Caucasian, Canada) LAL−01 (Caucasian, Canada) LAL−0008(Caucasian/Hispanic, LAL−06 (Caucasian/Hispanic, USA) USA) LAL−0004 (Caucasian, Canada) LAL−02 (Caucasian, Canada) LAL−0010(Hispanic, LAL−07 (Hispanic, USA/Chile) USA/Chile) LAL−0005 (Caucasian, Canada) LAL−03 (Caucasian, Canada) LAL−0015(Asian, LAL−10 (Asian,USA) USA) LAL−0006 (Caucasian, Germany) LAL−04 (Caucasian, Germany) LAL−0025(Caucasian/Asian, LAL−15 (Caucasian/Asian, USA) USA) LAL−0007 (Caucasian, USA) LAL−05 (Caucasian, USA) LAL−0058(Hispanic, LAL−31 (Hispanic, Colombia) Colombia) LAL−0012 (Caucasian, Germany/Belgium) LAL−32 LAL−0061(Hispanic, (Hispanic, Colombia) Colombia) 0.06 LAL−08 (Caucasian, Germany/Belgium) Present−day LAL−0013 (Caucasian, Spain) LAL−09 (Caucasian, Spain) European Populations LAL−0016 (Caucasian, Spain) LAL−11 (Caucasian, Spain) −Present−day HO Populations− LAL−0018 (Caucasian, USA) LAL−12 (Caucasian, USA) Central−South America LAL−0020 (Caucasian, Netherlands) LAL−13 (Caucasian, Netherlands) North America LAL−0022 (Caucasian, UK) LAL−14 (Caucasian, UK) Near−Middle East LAL−0028 (Caucasian, Canada) LAL−16 (Caucasian, Canada) Europe LAL−0029 (Caucasian, USA) LAL−17 (Caucasian, USA) Central−South Asia LAL−0030 (Caucasian, USA) 0.02 LAL−18 (Caucasian, USA) North Asia LAL−0032 (Caucasian, Germany) LAL−19 (Caucasian, Germany) East Asia LAL−0033 LAL−20 (Caucasian, (Caucasian,Italy) Italy) 0.04 LAL−0036 (Caucasian, Portugal) LAL−21 (Caucasian, Portugal) LAL−0039 (Caucasian, Germany LAL−22 (Caucasian, Germany LAL−0043 (Caucasian, Bulgaria) LAL−23 (Caucasian, Bulgaria) LAL−24 (Caucasian, LAL−0044 Netherlands) (Caucasian, Netherlands) LAL−0045 (Caucasian, Germany/France) LAL−25 (Caucasian, Germany/France) LAL−0046 (Caucasian, Germany) LAL−26 (Caucasian, Germany) LAL−0047 (Caucasian, USA/Canada) LAL−27 (Caucasian, USA/Canada) LAL−0049 (Caucasian, Canada) LAL−28 (Caucasian, Canada) LAL−0051 (Caucasian, Canada) LAL−29 (Caucasian, Canada) LAL−0055 (Caucasian, Canada) LAL−30 (Caucasian, Canada) 0.00 0.02 PC2 PC2 0.00 Present−day −0.02 Middle−Eastern Populations −0.02 −0.04 −0.04 −0.02 0.00 0.02 0.04 −0.02 −0.01 0.00 0.01 0.02 0.03 PC1 PC1 Figure S3. Principal Component Analysis (PCA) generated with the genomic data from the 1,980 West-Eurasian, Asian and Native American individuals genotyped in the Affymetrix Human origins (HO) array (Lazaridis et al. 2014). Related to Figure 2. A. Modern West-Eurasia populations in the Affymetrix HO array. B. Look-alike individuals for West-Eurasia projected in the Affymetrix HO array. C. Look-alike individuals for West-Eurasia, Asia and America projected in the Affymetrix HO array. A 0 2 B C 40 Xiong, et al., 2019 Hoskens, et al., 2021 Enrichment score 30 20 10 0 Seg1 Trait25 Seg1 Trait85 Seg2 Trait94 Seg2 Trait116 Prn-ChL Prn-ChR Seg3 Trait196 Seg3 Trait158 Figure S4. Heatmaps and Stratified LD score regression (S-LDSC) enrichment analysis. Related to Figure 2. A. Heatmap of hierarchical genetic clustering with bootstrap of genome-wide SNP genotyping arrays in the 16 look-alikes adding the reported face associated SNPs (Claes et al., 2018; Xiong et al., 2019; White et al., 2021; Facebase dataset; and GWAS central) to our 19,277 SNPs. Genotype clustering was performed using Euclidean distances and ward.D2 cluster method. Blue rectangles represent 11 look-alike pairs that unbiasedly clustered. 0 = homozygous reference SNPs (green), 1 = heterozygous SNPs (black) and 2 = homozygous alternate SNPs (red). B. S-LDSC enrichment estimates of selected facial traits for the 19,277 SNPs found to be shared in all “ultra” look-alike pairs. Facial traits with full summary statistics available from two studies (Xiong, et al., 2019 and Hoskens, et al., 2021) included in the GWAS catalog were analyzed. Only those traits with significant enrichment (enrichment score > 1 and enrichment p value < 0.05) are shown. Dashed line indicates enrichment score threshold. Bar colors indicate study of origin. Trait number indicates the features extracted per facial segment from Hoskens,et al., 2021. Prn: pronasale; ChL: left cheilion; ChR: right cheilion; Seg1: segment 1 (full face), Seg2: segment 2 (midface); Seg3 : segment 3 (outerface). C. Heatmap shows unsupervised genome-wide DNA methylation hierarchical clustering with bootstrap of the 9 “ultra” look-alikes, using all the methylation β-values obtained from MethylationEPIC arrays after Quality Check (QC) and preprocessing steps described in methods. A random selection of 5000 CpGs is represented. Colors represent a continuous quantification of methylation beta values at each CpG site where green highlights unmethylated CpG’s (0), black, 50% methylated CpG’s (0.5) and red, fully methylated CpG’s (1). Clustered look-alikes are shown in a blue rectangle. A B Figure S5. Alternative clustering and visualization methods for the DNA methylation data. Related to Figure 3. A. k- means clustering of the 16 look-alike couples using the Illumina HumanMethylation EPIC array data. Each color represents a cluster found by the k-means algorithm. B. t-distributed Stochastic Neighbor Embedding (t-SNE) visualization of the same DNA methylation data, using a random selection of 50,000 CpG sites. Each look-alike couple is depicted in a different color. A B 10.0 Age difference (years) 7.5 5.0 2.5 0.0 Non-Ultra Ultra look-alike look-alike 0 1 C D 0 55 0 0.5 Figure S6. DNA methylation age estimation, mEQTLs and microbiome analyses. Related to Figure 3. A. DNA methylation age estimation using the Hannum method for the 16 look-alike couples. Intrapair age estimation differences are significantly smaller (two-sided Mann-Whitney-Wilcoxon test P-value = 0.0052) among the 9 “ultra” look-alike pairs compared to the non-ultra look-alike pairs. B. Heatmap shows supervised genome-wide DNA methylation hierarchical clustering with bootstrap of the 16 look-alikes, using the 1,379 methylation β-values obtained from MethylationEPIC array after selecting all those CpGs located within a window of 100 base pairs from any of the 19,277 SNP’s found to be shared in all “ultra” look-alike pairs. Colors represent a continuous quantification of methylation beta values at each CpG site where green highlights unmethylated CpG’s (0), black, 50% methylated CpG’s (0.5) and red, fully methylated CpG’s (1) . Clustered “ultra” look-alikes are shown in a blue rectangle. C, D. Microbiome analysis of the 9 ultra” look-alikes. Heatmaps shows the distances from differences in pair-wise bacterial counts of species found in the microbiome of each look-alike pair (variation in alpha diversity scores) of counts from 0 – 55 (C) and relative proportions of the taxonomic profiles at the genus level (D) for each sample calculated on a scale of 0 - 0.5. Only the most represented genera are shown. Meta-genomic clustering of each look-alike sample was constructed using Euclidean distances and Ward.D2 hierarchical cluster method. Blue rectangle represents look-alike pairs whose microbiome is more closely related. A B C 20 Weight difference (Kg) 15 10 5 0 Non-Ultra Ultra look-alike look-alike Figure S7. Clustering for metagenomics species data and intrapair weight differences. Related to Figure 3. A, B. Alternative clustering and visualization methods for the metagenomics species data. A. k-means clustering of the 16 look- alike couples using the metagenomics species data. Each color represents a cluster found by the k-means algorithm. B. t- distributed Stochastic Neighbor Embedding (t-SNE) visualization of the same data. Each look-alike couple is depicted in a different color. C. Weight difference of the 16 look-alike couples. Intrapair weight differences are significantly smaller (two-sided Student’s t-test P-value = 0.035) among the 9 “ultra” look-alike pairs compared to the non-ultra look-alike pairs. Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people If for any reason you feel uncomfortable answering some of these questions, feel free to leave the space in white. To the best of our knowledge, do you have any familiar relationship with your look-alike: NO YES . If yes, specify __________________________ Name and Surname: _______________________________________________________ Gender: MALE FEMALE Race: CAUSASIAN HISPANIC AFRICAN ASIAN OTHER: _____ Age: ________ Date of birth: ___/___/___ (DD/MM/YY) Born in (location,country): _____________________________________________ Do you know how you were conceived? NO YES Natural conceive In Vitro Fertilization Current residence (locality and country): _______________________________ Location: CITY TOWN COUNTRYSIDE Did you live in other locations during your infancy / youth? Where?: Country: _____________ location: CITY TOWN COUNTRYSIDE Country: _____________ location: CITY TOWN COUNTRYSIDE Country: _____________ location: CITY TOWN COUNTRYSIDE BODY FEATURES: Weight: _________Kg Height: _________cm FACE FEATURES Hair: Natural color: WHITE BLOND RED-HAIRED BROWN BLACK BALD Natural shape: STRAIGHT WAVY CURLY Eyes: Color: GREY BLUE GREEN BROWN DARK Do you need glasses? NO SOMETIMES YES If "sometimes" or "yes", why? _______________________________ 1 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people PERSONAL FEATURES AND HABITS: Civil state: SINGLE Lived in couple before? NO YES t , time: ________ MARRIED/IN COUPLE Time you live in couple: _________________ DIVORCED/SEPARATED /WIDOWED Time you lived in couple: _______ DO YOU HAVE ANY DESCENDENCE? NO YES If "yes", how many?: ____ (nºsons: ___, nºdaughters: ___) What is the highest level of school you have completed or the highest degree you have received? Basic school High school University Others If "University": Degree: ________________ PhD: YES NO If "Others", which kind: ____________________________________ Employment status: EMPLOYEE AUTONOMOUS UNEMPLOYEE MEDICAL LEAVE RETIRED WORK INDOOR WORK OUTDOOR PHISICAL WORK OFFICE WORK TRAVEL WORK Kind of employment: Category: EXECUTIVE SALARIED OWN BUSSINESS What are your approximately yearly incomings? ___________________________ Do you have any pet(s)?: NO YES , If "yes": Kind of pet: ____________; Time: _________________ Do you have some significant alergies?: NO YES Gluten Lactics Nuts Antibiotics Dust/Polen Others: ___________________________________________________________ Blood group / RH: A B AB O / RH-positive RH-negative 2 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people Smoking status: NEVER FORMER LIGHT SMOKER HEAVY SMOKER If "former": Years since your last cigarrette: ______________ If "former","light","heavy": quantity (day/months) __________ Do you practice physical exercise?: NO YES Frequency: ___________ Indoor Outdoor Diet habits: No vegetarian Vegetarian Vegan Drinking habits: Alcohol: Never Sometimes Often Soft-drinks: Never Sometimes Often Other drinks : Coffee Tea Sometimes Often Have you suffered any significant illness? NO YES If yes, can you indicate which illness have you suffered? CHRONIC: Diabetes Chohn disease Others _______________ Cancer Heart failure Cerebral stroke Others _______________ Handwriting: Left-hand writer Right-hand writer 3 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people Si por alguna razón se siente incómodo respondiendo a algunas de estas preguntas, siéntase libre de dejar el espacio en blanco. Para nuestro mejor conocimiento, ¿tiene alguna relación familiar con su parecido?: NO SÍ En el caso "sí", especifique _____________________ Nombre y Apellidos: _____________________________________________________ Género: MASCULINO FEMENINO Raza: CAUCÁSICA HISPÁNICA AFRICANA ASIÁTICA OTRA: _____ Edad: ________ Fecha de nacimiento: ___/___/___ (DD/MM/YY) Nacido en (localidad, país): ___________________________________________ ¿Sabe usted cómo fue concebido? NO SÍ Concepción natural Fecundación In Vitro Residencia habitual (localidad, país): _______________________________ Localidad: CIUDAD PUEBLO EN EL CAMPO ¿Vivió en otros sitios durante su infancia/juventud? ¿Dónde?: País: _____________ localidad: CIUDAD PUEBLO CAMPO País: _____________ localidad: CIUDAD PUEBLO CAMPO País: _____________ localidad: CIUDAD PUEBLO CAMPO CARACTERÍSTICAS CORPORALES: Peso: _________Kg Altura: _________cm RASGOS FACIALES Cabello: Color natural: BLANCO RUBIO PELIROJO MARRÓN NEGRO CALVO Forma natural: LISO ONDULADO RIZADO CALVO Ojos: Color: GRIS AZUL VERDE MARRÓN OSCURO Tamaño: PEQUEÑO MEDIANO GRANDE ¿Necesita gafas? NO A VECES SÍ Si "a veces" o "sí", por qué? _______________________________ 4 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people CARACTERÍSTICAS PERSONALES Y HÁBITOS: Estado civil: SOLTER@ Ha vivido en pareja? NO SÍ , tiempo: ________ CASAD@/EN PAREJA Tiempo que ha vivido en pareja:___________ DIVORCIAD@/SEPARAD@/VIUD@ Tiempo que vivió en pareja:__________ HA TENIDO USTED DESCENDENCIA? NO SÍ En caso "sí", ¿cuánta?: ____ (nº hijos: ___, nº hijas: ___) ¿Cuál es el nivel más alto de estudios que ha completado o el grado más alto que ha recibido? Escuela básica Instituto Universidad Otros If "Universidad": Licenciatura: _________Doctorado: SÍ NO If "Otros", ¿qué tipo?: ____________________________________ Estado laboral: EMPLEAD@ AUTÓNOM@ DESEMPLEAD@ BAJA MÉDICA JUBILAD@ TRABAJO EN INTERIOR EN EL EXTERIOR TRABAJO FÍSICO DE OFICINA VIAJANDO Clase de empleo: Categoría: EJECUTIV@ ASALARIAD@ NEGOCIO PROPIO ¿Cuánto son sus ingresos anuales aproximados? ___________________________ ¿Tiene alguna mascota(s)?: NO SÍ , En caso "sí": Qué clase: _______________; Tiempo: ___________ ¿Tiene alguna alérgia(s) importante(s)?: NO SÍ Glúten Lácticos Frutos secos Antibióticos Polvo/Pólen Otras: ___________________________________________________________ Grupo sanguíneo / RH: A B AB O / RH-positivo RH-negativo 5 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people Tabaquismo: NUNCA EXFUMADOR/A FUMADOR/A LIGER@ FUMADOR/A SEVER@ Si "exfumador/a": Años des de su último cigarrillo:__________ Si "ex", "liger@","sever@": cantidad (día/mes) __________ ¿Practica usted ejercicio físico?: NO SÍ Frecuencia: _______ Interior Exterior Hábitos dietéticos: No vegetarian@ Vegetarian@ Vegan@ Hábitos de bebida: Alcohol: Nunca A veces A menudo Refrescos: Nunca A veces A menudo Otras bebidas: Café Té A veces A menudo ¿Ha sufrido alguna enfermedad importante? NO SÍ En caso "sí", ¿podría usted indicar cuál(es)? CRÓNICA: Diabetes Chron Otra(s) ___________________ Cáncer Fallo cardíaco Ictus Otra(s _____________ Escritura: Zurd@ Diestr@ 6 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people Si pour une raison ou pour une autre vous ne souhaitez pas communiquer certaines informations, merci de laisser libre l’espace correspondant. Nom et Prénom : _______________________________________________________ A votre connaissance, avez-vous un lien de parenté avec votre sosie ? NO YES Si oui, précisez : __________________________ Genre : Homme Femme Ethnie : Caucasien Hispanique Aficain Asiatique Autre : _____ Age : ________ Date de naissance : ___/___/___ (j/ms/an) Lieu de naissance (localité,pays): _____________________________________________ Est-ce que vous savez comment vous avez été conçus ? NON OUI Conception naturelle Fécondation in vitro Lieu actuel de résidence (localité, pays) : _______________________________ Localité : Ville Village Campagne Avez-vous vécu dans un autre endroit durant votre enfance/jeunesse ? Pays : ___________ localité: Ville Village Campagne Pays : ___________ localité: Ville Village Campagne Pays : ___________ localité: Ville Village Campagne 7 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people Mesures corporelles : Poids : _________Kg Taille : _________cm Caractéristiques du visage Cheveux : Couleur naturelle : BLANC BLOND ROUX MARRON NOIR CHAUVE Allure naturelle : RAIDE ONDULÉ BOUCLÉ Yeux : Couleur : GRIS BLEU VERT MARRON NOIR Portez-vous des lunettes ? NON PARFOIS OUI Si “oui” ou “parfois”, précisez ______________________________ Caractéristiques personnelles et loisirs : Etat civil: Célibataire Avez-vous vécu en couple auparavant ? NON OUI Combien de temps ? : ________ Marié/En couple Depuis combien de temps vivez-vous en couple ? _________________ DIVORCÉ/SÉPARÉ/VEUF/VEUVE Depuis combien de temps viviez-vous en couple ? : _______ AVEZ-VOUS DES ENFANTS ? NON OUI Si “oui”, combien : ____ (nb de fils : ___, nb de filles: ___) Quel le plus haut niveau d’étude scolaire ou le diplôme le plus élevé que vous avez atteint ? Collège Lycée Université Autre Si "Université” : Niveau : ________________ PhD: OUI NON Si "Autre", précisez : ____________________________________ 8 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people Emplois : Employé Indépendant Sans emploi Congés maladie Retraité PAS DE TRAVAIL EN EXTERIEUR TRAVAIL EN EXTERIEUR TRAVAIL PHYSIQUE TRAVAIL DE BUREAU DEPLACEMENT FREQUENT TYPE DE TRAVAIL : Catégorie : DIRIGEANT SALARIÉ ENTREPRENEUR Quels sont approximativement vos revenus annuels ? ___________________________ Avez-vous un ou plusieurs animaux domestiques : NON OUI Si "oui” : quel(s) animal(s) ?: ____________ Combien de temps ?: _________________ Souffrez-vous d’allergies gênantes ? NON OUI Gluten Lactose Noix Antibiotique Poussière/pollen Autre(s):__________________________________________________________ Groupe sanguin / Rhésus : A B AB O Rhésus-positif Rhésus-négatif Tabagisme : Non-fumeur Sevré Petit fumeur Fumeur régulier Si "sevré” : date de sevrage (année) : ______________ Si "sevré”, "petit fumeur”, "fumeur régulier” : nombre de cigarettes par jour ou mois : __________ Pratiquez-vous une activité physique ? NON OUI Fréquence : ___________ Intérieur Extérieur Habitudes alimentaires : Non végétarien Végétarien Végane 9 Methods S1: Data collection questionnaire, DATA COLLECTION related to STAR Methods QUESTIONNAIRE Project: Genomic characterization to study common features of look-alike people Habitudes de boisson : Alcool : Jamais Parfois Régulièrement Soda : Jamais Parfois Régulièrement Autres boissons : Café Thé Parfois Régulièrement Avez-vous présenté des maladies ayant impactées ou impactant significativement votre quotidien ? NO YES Si oui, pouvez-vous préciser de quelle(s) maladie(s) s’agit-il ? Chronique : Diabète Maladie de Crohn Autres _______________ Cancer Insuffisance cardiaque Accident vasculaire cérébral Autres _______________ Main dominante : Gaucher Droitier 10