New Publication in Life Sciences: predicting the effect of a drug
Which drug candidate will be most effective for a particular disease? Such predictions are processed on computers using machine learning and other artificial intelligence approaches. Often it is a question of how well a compound binds to a target protein and thus triggers a specific chain of action in the body.
The b-it scientists Jannik P. Roth and Prof. Dr Jürgen Bajorath have advanced the use of machine learning (ML) in drug discovery. They explored how ML models predict drug efficacy by binding to target proteins. The study, published in "Cell Reports Physical Science," reveals that different ML model variants can produce similar predictions but with vastly different underlying explanations. By applying the Shapley value concept from game theory, they quantified the contributions of molecular features to predictions, highlighting the challenge of interpreting these results consistently.
For more details, visit University of Bonn News.
Publication:
Jannik P. Roth, Jürgen Bajorath. Machine Learning Models with Distinct Shapley Value Explanations for Chemical Compound Predictions Decouple Feature Attribution and Interpretation, Cell Reports Physical Science, DOI: 10.1016/j.xcrp.2024.102110, URL: https://doi.org/10.1016/j.xcrp.2024.102110
