Bison improves drug response prediction across diverse datasets
Bison: Cross-Dataset Learning for Unseen-Compound Perturbation Prediction
Machine LearningArtificial Intelligence
Summary
Predicting how genes react to new drugs is hard because data comes from different experiments using different chemicals and gene tests. The authors show that even when overall gene responses look similar, predicting specific drug effects is still difficult. They created Bison, which learns from multiple datasets by connecting gene information and focusing on drug differences. Bison better predicts responses to new drugs compared with other methods, even when trained on all datasets together. This approach helps use varied chemical data more effectively to understand unseen drug effects.
What this means in practice
- •For drug discovery teams: Predict gene responses to novel compounds using integrated data from multiple chemical screening experiments.
- •For bioinformatics software developers: Build tools that improve drug response prediction by combining diverse datasets with shared gene representations and drug contrast learning.
Authors
Yunfan Liu, Kasra Ghorbani, Yufei Huang, Zicheng Liu, Jiangbin Zheng, Jingbo Zhou, Shaorong Chen, Chang Yu, Stan Z. Li
Abstract
Predicting transcriptional responses to unseen compounds is limited by fragmented chemical coverage and heterogeneous experimental platforms and gene panels. To assess molecular generalization across these settings, we build on Chem-PerturBridge to benchmark eight datasets with 16,771 compounds, withholding test compounds from every training dataset. This comparison reveals that high overall response agreement can coexist with weak prediction of drug-specific differences, despite reproducible signals across repeated measurements. To exploit complementary chemical supervision while targeting these differences, we introduce Bison: a shared gene representation connects native panels, while two discrete diffusion models compose context-dependent responses with molecular deviations learned through matched drug-contrast supervision. A single Bison model jointly trained across all eight datasets achieves the highest mean overall-response and drug-contrast Pearson correlations on the full benchmark in comparison with 11 methods trained independently per dataset. Compared with dataset-specific training of the same architecture, joint training increases mean drug-contrast correlation by 27.4\%, with gains across all eight datasets and improvements in overall response prediction. These results demonstrate how matched drug contrasts turn complementary screens into shared molecular supervision for unseen-drug response prediction while preserving native gene measurements.