Supplementary Material for: Early pseudoprogression and progression lesions in glioblastoma patients are both metabolically heterogeneous Gülnur S. Ungan 1,3, Albert Pons-Escoda 2, Daniel Ulinic 3, Carles Arús 3,1, Sandra Ortega- Martorell 4, Iván Olier 4, Alfredo Vellido 5,1, Carles Majós 2,1, Margarida Julià-Sapé 3,1* 1 Centro de Investigación Biomédica en Red (CIBER), Spain 2 Grup de Neuro-oncologia, Institut d'Investigació Biomèdica de Bellvitge (IDIBELL), Hospital Universitari de Bellvitge, Spain 3 Departament de Bioquímica i Biologia Molecular and Institut de Biotecnologia i Biomedicina (IBB), Universitat Autònoma de Barcelona (UAB), Spain 4 School of Computer Science and Mathematics, Liverpool John Moores University (LJMU), United Kingdom 5 IDEAI-UPC research center, UPC BarcelonaTech, Spain 1 Abbreviations list SCER: suspicious contrast-enhancing voxel, PP: pseudoprogression, R: response, S: stable, TP: true progression 2 Patients Summary of patients and voxels used per patient. Supplementary table 1. NEV: no enhancing voxels; EXCL: excluded due to lack of SCER in the P1M exploration; NoExp: Patient was not explored. % of Number Fulfils Number of SCER Number of SCER Number criteria Number artifactual artifactual MGMT WHO of SCER voxels of SCER IDH status Date P1M 3 Label of voxels SCER voxels Patient metylation classification voxels at included voxels > months in grid voxels excluded P1M in the SNR 20 post RX (excluded) from analysis analysis 10 Unavailable YES 2007 14/07/2016 YES TP 256 2 2 2 0 0 11 Negative YES 2016 28/07/2016 YES TP 256 1 1 1 0 0 12 Unavailable YES 2007 09/09/2016 YES TP 256 7 7 5 0 0 13 Negative YES 2016 06/10/2016 YES PP 256 13 13 12 0 0 14 Negative YES 2016 18/10/2016 YES PP 256 2 2 0 0 0 15 Negative YES 2016 25/10/2016 YES R 256 4 4 4 0 0 16 Negative NO 2016 18/11/2016 YES TP 256 12 12 12 0 0 17 Negative YES 2016 10/01/2017 YES S 256 9 8 7 1 11,11 18 Negative YES 2016 02/02/2017 NEV R 256 19 Negative YES 2016 21/02/2017 YES PP 256 1 0 0 1 100 20 Negative NO 2016 24/02/2017 YES R 256 4 0 0 4 100 21 Negative YES 2016 10/03/2017 YES PP 256 11 9 0 2 18,18 22 Negative YES 2016 10/03/2017 YES TP 256 4 0 0 4 100 23 Negative NO 2016 16/03/2017 YES PP 256 6 0 0 6 100 24 Negative NO 2016 13/04/2017 YES PP 256 16 12 10 4 25 25 Negative NO 2016 13/04/2017 YES R 256 4 2 2 2 50 26 Negative YES 2016 26/04/2017 YES PP 256 5 5 5 0 0 27 Negative YES 2016 12/05/2017 YES PP 256 10 1 1 9 90 28 Negative NO 2016 30/05/2017 YES TP 256 42 0 0 42 100 29 Negative Unknown 2016 15/06/2017 NO EXCL 256 30 Negative NO 2016 11/07/2017 YES TP 256 8 8 7 0 0 31 Negative NO 2016 21/07/2017 YES PP 256 22 17 12 5 22,73 2 % of Number Fulfils Number of SCER Number of SCER Number criteria Number artifactual artifactual MGMT WHO of SCER voxels of SCER IDH status Date P1M 3 Label of voxels SCER voxels Patient metylation classification voxels at included voxels > months in grid voxels excluded P1M in the SNR 20 post RX (excluded) from analysis analysis 32 Negative YES 2016 10/08/2017 YES TP 256 7 7 7 0 0 33 Negative NO 2016 10/08/2017 NEV S 256 34 Negative NO 2016 23/08/2017 YES TP 256 12 7 5 5 41,67 35 Negative YES 2016 30/08/2017 YES PP 256 7 5 4 2 28,57 36 Negative NO 2016 05/09/2017 NEV S 256 37 Negative NO 2016 26/09/2017 YES PP 256 3 3 3 0 0 38 Negative YES 2016 28/09/2017 NoExp EXCL 256 39 Positive YES 2016 11/10/2017 YES TP 256 13 13 13 0 0 40 Negative NO 2016 14/11/2017 NEV R 256 41 Negative NO 2016 21/11/2017 YES R 256 8 6 6 2 25 42 Negative YES 2016 30/11/2017 YES PP 256 6 6 3 0 0 43 Negative NO 2016 03/01/2018 NEV S 256 44 Negative NO 2016 03/01/2018 YES TP 256 11 11 10 0 0 45 Negative YES 2016 23/01/2018 NEV S 256 46 Negative NO 2016 08/03/2018 YES PP 256 10 3 2 7 70 47 Negative NO 2016 22/03/2018 YES PP 256 11 11 3 0 0 48 Negative YES 2016 22/03/2018 NEV S 256 49 Negative NO 2016 27/03/2018 YES TP 256 3 2 1 1 33,33 50 Negative YES 2016 12/04/2018 Exitus EXCL 256 51 Negative NO 2016 12/04/2018 YES TP 256 21 17 9 4 19,05 52 Negative NO 2016 12/04/2018 YES PP 256 17 0 0 17 100 53 Negative YES 2016 15/05/2018 YES S 256 16 3 0 13 81,25 54 Negative YES 2016 25/06/2018 YES S 256 7 7 1 0 0 55 Negative NO 2016 02/07/2018 YES TP 256 4 0 0 4 100 56 Negative NO 2016 21/09/2018 YES PP 256 70 11 11 59 84,29 57 Negative NO 2016 28/11/2018 NEV S 256 58 Negative NO 2016 28/11/2018 YES PP 256 7 4 1 3 42,86 59 Negative YES 2016 11/01/2019 YES PP 256 28 14 9 14 50 60 Negative NO 2016 21/01/2019 YES PP 256 3 3 0 0 0 61 Negative NO 2016 21/01/2019 YES PP 256 3 3 3 0 0 62 Negative NO 2016 25/01/2019 YES PP 256 5 5 1 0 0 Total 13568 455 244 10752 3 3 Classifier parameters * • TP vs PP • LDA {'n_components': 0, 'solver': 'svd'} Number of components (< n_classes - 1) for dimensionality reduction. Solver: solver to use. • LR {'C': 0.1, 'penalty': 'l2'} C: Inverse of regularization strength Penalty: Specify the norm of the penalty • RF {'max_features': 'sqrt', 'n_estimators': 50} Max_features: The number of features to consider when looking for the best split N_estimators: The number of trees in the forest • SVM {'C': 100, 'gamma': 0.01} C: Inverse of regularization strength Gamma: Kernel coefficient • TP vs PP+R+S a. LDA {'n_components': 0, 'solver': 'svd'} Description: ‘n_components’: Number of components (< n_classes - 1) for dimensionality reduction. 'solver': Solver: solver to use. b. LR {'C': 1.0, 'penalty': 'l2'} ‘C’: Inverse of regularization strength ’penalty’: Specification of the norm of the penalty c. RF {'max_features': 'sqrt', 'n_estimators': 50} ´max_features’: The number of features to consider when looking for the best split ’n_estimators’: The number of trees in the forest d. SVM {'C': 1, 'gamma': 1} ‘C’: Inverse of regularization strength ’gamma’: Kernel coefficient * Software used for the classifiers, available from: https://scikit-learn.org/stable/index.html 4 4 Classification results for two classes (TP vs {PP, R and S}) Supplementary table 2. Classifier Train/Test Classification Balanced Accuracy Sensitivity Specificity AUC task Accuracy LDA Train Patient 70.1 71.2 ± 3.3 65.7 ± 3.3 75.3+/-3.2 70.1 ± 6.4 LDA Test Patient 66.3 66.4 ± 3.3 66.4 ± 3.3 75.3 ± 3.5 66.4 ± 3.3 LR Train Patient 70.4 75.2 ± 2.2 65.1 ± 2.8 92.2 ± 2.7 70.3 ± 3.0 LR Test Patient 68.3 63.4 ± 2.2 59.7 ± 2.8 79.8 ± 2.1 70.1 ± 2.2 RF Train Patient 68.7 75.2 ± 3.5 33.1± 3.5 79.9 ± 3.2 68.1 ± 3.1 RF Test Patient 55.4 56.8 ± 5.6 33.3 ± 6.4 74.7 ± 4.3 54.4 ± 3.7 SVM Train Patient 72.4 75.2± 3.5 75.2± 2.1 65.6 ± 4.7 67.9 ± 3.2 SVM Test Patient 70.1 71.4 ± 3.5 75.3± 3.5 66.2 ± 3.3 70.4 ± 3.2 LDA Train Voxel 69.2 67.3 ± 2.3 77.8 ± 2.2 53.5 ± 2.2 65.7± 2.2 LDA Test Voxel 68.3 65.6 ± 2.2 78.1 ± 2.4 54.1 ± 2.4 65.3 ± 2.2 LR Train Voxel 68.7 70.1± 2.7 60.2± 3.7 92.1± 2.1 70.1 ± 2.7 LR Test Voxel 65.2 58.3 ± 3.3 62.1± 2.7 90.1 ± 1.1 64 ± 3.5 RF Train Voxel 65.1 71.2 ± 2.2 50.1 ± 2.5 80.1± 1.8 65.4 ± 2.2 RF Test Voxel 63.3 68.2 ± 2.2 45.4± 2.4 78.2 ± 1.8 62.3 ± 2.1 SVM Train Voxel 67.2 69.1± 2.8 78.2 ± 2.3 48.2 ± 2.5 66.4 ± 2.1 SVM Test Voxel 69.3 67.1± 2.1 78.4 ± 2.1 54.2 ± 2.1 66.2 ± 1.9 Train/Test Split: 80/20. Balanced accuracy, accuracy, sensitivity, specificity and AUC in % ± 95% CI. 5 Superimposed spectra of voxels, not taking into account any SNR Threshold Supplementary figure 1. Overlay of individual spectra of each class and mean (thick line). 5 6 Source extraction results Supplementary figure 2. The three different solutions obtained when extracting four sources. Supplementary figure 3. Four different solutions obtained when extracting ten sources. 6 Supplementary figure 4. One solution obtained when extracting twenty sources. Supplementary figure 5. Another solution when extracting twenty sources. Supplementary figure 6. Another different solution obtained when extracting twenty sources. 7 7 Color maps for all patients included in the analysis For all figures from 7 to 37, the legend is the same. From left to right: source 1 contribution map (yellow), source 2 contribution map (green), source 3 contribution map (red) and winning source map, from the SCER voxels, overlaid onto the reference image. 8 Supplementary figure 7. Case 10: TP. Supplementary figure 8. Case 11: TP. Supplementary figure 9. Case 12: TP. Supplementary figure 10. Case 13: PP. 9 Supplementary figure 11. Case 15: R. Supplementary figure 12. Case 16: TP. Supplementary figure 13. Case 17: S. Supplementary figure 14. Case 24: PP. 10 Supplementary figure 15. Case 25: R. Supplementary figure 16. Case 26: PP. Supplementary figure 17. Case 27: PP. Supplementary figure 18. Case 30: TP. 11 Supplementary figure 19. Case 31: PP. Supplementary figure 20. Case 32: TP. Supplementary figure 21. Case 34: TP. Supplementary figure 22. Case 35: PP. 12 Supplementary figure 23. Case 37: PP. Supplementary figure 24. Case 39: TP. Supplementary figure 25. Case 41: R. Supplementary figure 26. Case 42: PP. 13 Supplementary figure 27. Case 44: TP. Supplementary figure 28. Case 46: PP. Supplementary figure 29. Case 47: PP. Supplementary figure 30. Case 49: TP. 14 Supplementary figure 31. Case 51: TP. Supplementary figure 32. Case 54: S. Supplementary figure 33. Case 56: PP. Supplementary figure 34. Case 58: PP. 15 Supplementary figure 35. Case 59: PP. Supplementary figure 36. Case 61: PP. Supplementary figure 37. Case 62: PP. 16