The physics-AI dialogue in drug design

Area of Science:

• Biochemistry
• Computational Chemistry
• Drug Discovery

Background:

• Protein structure determination has evolved significantly since 1960.
• Machine learning (ML) methodologies in protein science were recognized with the 2024 Nobel Prize in Chemistry.
• ML applications are increasingly integrated into drug discovery workflows.

Purpose of the Study:

• To review the evolution and current state of ML in protein science and drug discovery.
• To highlight challenges and emerging applications of ML.
• To emphasize the synergy between ML and physics-based methods.

Main Methods:

• Review of machine learning techniques in protein structure prediction and design.
• Analysis of ML applications in drug discovery, including pose scoring and molecular descriptor generation.
• Discussion of the complementarity between ML and physics-based methods like molecular dynamics simulations.

Main Results:

• ML tools show promise in interpolating between compounds for hit-to-lead optimization.
• Physics-based methods, such as free energy calculations, appear superior for novel derivative design.
• A growing synergy exists between physics-based and ML techniques.

Conclusions:

• ML has revolutionized protein science, but challenges remain in predicting structural ensembles.
• Balancing ML's potential benefits with environmental costs is crucial.
• Interdisciplinary understanding of ML's advantages and limitations is essential for exploiting its synergy with physics-based methods in drug design.