PSnpBind-ML: predicting the effect of binding site mutations on protein-ligand binding affinity

Area of Science:

• Computational biology
• Pharmacogenomics
• Machine learning in drug discovery

Background:

• Protein mutations, particularly in binding sites, significantly influence individual drug responses, affecting efficacy and side effects.
• Experimental screening of ligand-protein binding against variants is costly and time-consuming, necessitating efficient in silico methods.
• Existing computational approaches for predicting mutation effects often rely on limited datasets.

Purpose of the Study:

• To develop and validate a machine learning model for predicting protein-ligand binding affinity changes due to single-point mutations in binding sites.
• To leverage a large-scale dataset from extensive docking experiments to train robust predictive models.
• To provide a tool for early-stage drug discovery to assess the impact of population-level protein variants on drug binding.

Main Methods:

• Utilized the PSnpBind database encompassing six hundred thousand docking experiments.
• Developed a machine learning approach with two regression models: one for wild-type and one for mutated protein-ligand binding affinity.
• Encoded protein, binding site, mutation, and ligand information using 256 features, with half selected based on domain knowledge.

Main Results:

• Achieved high prediction accuracy on an independent test set, with Root Mean Square Error (RMSE) between 0.5–0.6 kcal/mol and R-squared (R²) values ranging from 0.87–0.90.
• Demonstrated improved prediction performance compared to several previously reported models for protein-ligand binding affinity.
• Successfully predicted binding affinity for both wild-type and mutated protein-ligand complexes.

Conclusions:

• The developed machine learning models offer a computationally efficient method to predict the impact of protein mutations on drug binding affinity.
• These models can serve as a valuable complementary tool in early drug discovery, guiding experimental efforts.
• The approach facilitates rapid assessment of ligand binding affinity across diverse protein variants, aiding in the identification of broadly effective drug candidates.