Synergy and Complementarity between Focused Machine Learning and Physics-Based Simulation in Affinity Prediction.

Area of Science:

• Computational chemistry
• Drug discovery

Background:

• Ligand affinity prediction is crucial for drug discovery.
• Physics-based simulation (FEP+) and machine learning (QuanSA) are established methods.
• Assessing the complementarity of these approaches is key to optimizing prediction accuracy.

Purpose of the Study:

• To evaluate the synergy between FEP+ and QuanSA for ligand affinity prediction.
• To determine if a hybrid approach offers superior performance over individual methods.
• To test the generalizability of this synergy across diverse biological targets.

Main Methods:

• Applied FEP+ and QuanSA to predict activity of LFA-1 inhibitors.
• Developed a hybrid model by averaging predictions from FEP+ and QuanSA.
• Validated the hybrid model on two public FEP+ benchmarks with 16 diverse targets.

Main Results:

• Both FEP+ and QuanSA provided accurate predictions within their domains.
• The hybrid model significantly improved ranking and absolute pKi values compared to individual methods.
• Accurate QuanSA models were derived using focused training data on relevant ligands.

Conclusions:

• FEP+ and QuanSA are complementary for ligand affinity prediction.
• A hybrid approach combining physics-based and machine learning methods enhances prediction accuracy.
• Careful application of each method and hybrid strategies are recommended for optimal affinity prediction.