AI for Healthcare

Summary

Artificial intelligence can help connect complex biomedical data to useful predictions, but high benchmark performance does not guarantee that a model will generalize to new molecules, populations or clinical settings. Our research examines both the opportunities and the limits of machine learning in healthcare.

We combine AI with network-based representations to improve prediction, interpretability and evaluation. The emphasis is on methods that reveal the evidence behind a prediction, avoid shortcuts hidden in biomedical datasets and remain useful beyond well-annotated examples.

Research directions

• Generalizable protein-ligand and drug-target prediction
• Network-aware machine learning
• Interpretable models for biomedical data
• Evaluation of bias, shortcuts and external validity
• Integration of molecular, clinical and population evidence

Selected publication

• Chatterjee A, Walters R, Shafi Z, et al. Improving the generalizability of protein-ligand binding predictions with AI-Bind. Nature Communications (2023). doi:10.1038/s41467-023-37572-z