// Cancer Drug Resistance Prediction Engine  ·  458 drugs  ·  978 cell lines  ·  threshold=0.51
Extra Trees  AUC 0.9225
Random Forest  AUC 0.962
XGBoost  AUC 0.9295
Ensemble prediction from Extra Trees + Random Forest + XGBoost trained on 1.7M GDSC + DepMap records. Classification threshold: 0.51 (auto-optimised for F1).
🔬
Drug × Cell Line Resistance Prediction
Running ensemble models...
Model Scores — Resistance Probability
Extra Trees
Random Forest
XGBoost
Ensemble
Feature Explanation (SHAP)
Computing SHAP values...
🧬
Cell Line Drug Resistance Profile
Computing 458 drug predictions...
🧬
Select a cell line to see resistance profile across all 458 drugs
💊
Drug Resistance Across Cell Lines
Computing 978 cell line predictions...
💊
Select a drug to see resistance pattern across all 978 cell lines
🧠
Global Feature Importance (SHAP)
SHAP (SHapley Additive exPlanations) shows which biological and chemical features drive resistance predictions globally across the dataset. Higher values = more influence on predictions.
Top 20 Most Influential Features
drug_enc
0.3954
tpsa
0.3292
alogp
0.2599
molwt
0.2512
hbd
0.1947
rot_bonds
0.1136
subtype_PC1
0.0891
hba
0.0779
expr_PC2
0.0776
expr_PC7
0.0449
cell_enc
0.0392
cnv_PC7
0.0388
geneEffect_PC1
0.0364
expr_PC4
0.0361
LoHFraction
0.0343
mut_PC6
0.0307
geneEffect_PC3
0.0298
dep_PC7
0.0283
expr_PC8
0.0259
subtype_PC6
0.0254
🔍
Per-Prediction SHAP Explanation
Select a drug and cell line to see exactly which features pushed this specific prediction toward resistance or sensitivity.
Computing SHAP values...
📊
Model Performance
Training: Full 1,756,307 records — GDSC1 + GDSC2 drug-response integrated with DepMap multi-omics. Resistance threshold: 0.51 (auto-optimised for F1 score). Test set: 20% stratified holdout.
ModelROC-AUCF1 ScoreAccuracyAUC Visual
Extra Trees 0.9225 0.87 0.8349
Random Forest 0.962 0.9085 0.8917
XGBoost 0.9295 0.8718 0.8488
Ensemble 0.958 0.904 0.8856
Feature Groups
Drug Chemistry
MolWt, TPSA, AlogP, HBA, HBD, RotBonds (RDKit descriptors)
Transcriptomics
Expression PCs 1–8 (~95% variance, DepMap RNA-seq)
Mutations
Somatic mutation PCs 1–8 (damaging variants)
Copy Number
CNV PCs 1–8 (gene-level log₂)
CRISPR Screens
Gene effect + dependency PCs 1–8
Genomic Signatures
MSIScore, Ploidy, CIN, WGD, LoHFraction, Aneuploidy
Cancer Subtypes
Subtype PCs 1–8 + 20 lineage flags
Gene Fusions
10 high-confidence fusion pair flags