Endometrial Cancer Molecular Features prediction and Visualization Using Deep Learning

https://doi.org/10.1016/j.xcrm.2021.100400

Features included

• 18 mutations (ARID1A, ATM, BRCA2, CTCF, CTNNB1, FAT1, FBXW7, FGFR2, JAK1, KRAS, MTOR, PIK3CA, PIK3R1, PPP2R1A, PTEN, RPL22, TP53, ZFHX3)
• 4 molecular subtypes (CNV.H, CNV.L, MSI, POLE)
• Histological subtypes (Endometrioid, Serous)

Architecture included

• Inception (V1, V2, V3, V4)
• inception-ResNet (V1, V2)
• Panoptes(X & F) (V1, V2, V3, V4)

Catalogue of codes including all statistical analyses codes

• Accessory.py: Accessory functions for Inception Models, including AUROC/AUPRC plotting, CAM, etc.
• Accessory2.py: Accessory functions for X Models, including AUROC/AUPRC plotting, CAM, etc.
• annotation_plot.R: Build data summary heatmap.
• Cutter.py: Bulk cutting svs images and normalization.
• Cutter_NYU.py: Bulk cutting svs images and normalization for NYU samples.
• cnn4.py: Tensorflow driving code for single resolution model training/validation/testing.
• cnn5.py: Tensorflow driving code for multi resolution model training/validation/testing.
• data_input2.py: Reading TFrecords file for single resolution tiles.
• data_input3.py: Reading TFrecords file for multi resolution tiles.
• data_input_fusion.py: Reading TFrecords file for multi resolution tiles with BMI and age.
• Fusion_prep.py: Label preparation with BMI and age integrated.
• HE_mosaic.py: Make tSNE mosaic plots (single resolution).
• HE_mosaic2.py: Make tSNE mosaic plots (multi resolution).
• Img_summary.R: Summary of images in the cohort (dimension, # images per patient, counts).
• InceptionV1.py: InceptionV1 architecture.
• InceptionV2.py: InceptionV2 architecture.
• InceptionV3.py: InceptionV3 architecture.
• InceptionV4.py: InceptionV4 architecture.
• InceptionV5.py: InceptionResnetV1 architecture.
• InceptionV6.py: InceptionResnetV2 architecture.
• Label_prep2.py: Label preparation.
• mainm3.py: Main method for single resolution model training/validation/testing.
• mainm4.py: Main method for multi resolution model training/validation/testing.
• make_table.R: Summarize all the prediction tasks results in a table and a heatmap.
• MW_test.R: Tile-level Wilcoxon tests and plotting.
• Model_stat_test.R: Patient-level Wilcoxon tests, t-tests, AUROC tests, and plotting.
• multi_stat_test.R: compare multi-resolution and single resolution models.
• NYU_data_prep.py: NYU data preparation.
• NYU_loaders.py: Loading NYU data.
• NYU_test.py: Run testing on NYU dataset.
• POLE_pred.R: Multi-model systems to predict POLE subtype.
• RGB_profiler.py: Get RGB summary of tiles in cohort.
• RealtestV4.py: Deployment code for trained models.
• Realtest_for_figure.py: Deployment code for best performing trained models.
• ROC_figure.R: Example ROC plot for figures.
• Sample_prep.py: Sample preparation code for single resolution models, including sampling.
• Sample_prep2.py: Sample preparation code for multi resolution models, including sampling.
• scatter_logits.R: Plot prediction logits for figure 6.
• Similarities.R: YuleY similarity calculations between features.
• Slicer.py: Multi-thread cutting of images.
• Slicer_NYU.py: Multi-thread cutting of NYU images.
• Slide_Size_Count.py: Image dimension summary and number of tiles per image counts.
• Statistics_MSI.R: Statistical metrics for MSI predictions.
• Statistics_histology.R: Statistical metrics for histological subtype predictions.
• Statistics_mutations.R: Statistical metrics for mutation predictions.
• Statistics_special.R: Statistical metrics for other types of predictions.
• Statistics_subtypes.R: Statistical metrics for molecular subtype predictions.
• Statistics_NYU.R: Statistical metrics for NYU samples.
• Summary.R: Count number of patients for each task in the cohort.
• SummaryTable.R: Summary table for the paper.
• tSNE.R: tSNE dimensional reduction for activation maps.
• tSNE_for_figure.R: High quality tSNE dimensional reduction for activation maps.
• UMAP.R: UMAP dimensional reduction for activation maps.
• X1.py: Panoptes2 architecture.
• X2.py: Panoptes1 architecture.
• X3.py: Panoptes4 architecture.
• X4.py: Panoptes3 architecture.
• Legacy: Deprecated codes.