SMPLIP-Score: predicting ligand binding affinity from simple and interpretable on-the-fly interaction fingerprint...

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Machine learning in bioinformatics

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for drug discovery lead optimization.
• Traditional scoring functions often lack collinearity with experimental binding data.
• Current machine learning models, while accurate, suffer from complex and uninterpretable featurization processes.

Purpose of the Study:

• To develop an interpretable predictor for absolute binding affinity.
• To overcome the limitations of complex featurization in existing machine learning models.
• To introduce SMPLIP-Score (Substructural Molecular and Protein-Ligand Interaction Pattern Score) as a novel solution.

Main Methods:

• Developed SMPLIP-Score with simple featurization embedding ligand-binding site environment and ligand molecular fragments.
• Utilized a vectorized matrix input for learning layers, including random forest and deep neural networks.
• Evaluated model performance on benchmark datasets: PDBbind v.2015, Astex Diverse Set, CSAR NRC HiQ, FEP, PDBbind NMR, and CASF-2016.

Main Results:

• SMPLIP-Score achieved a Pearson's correlation coefficient up to 0.80 and a root mean square error up to 1.18 in pK units.
• Demonstrated comparable performance to state-of-the-art models despite less complex features.
• Confirmed generality, predictive power, ranking power, and robustness through direct interpretation of feature matrices.

Conclusions:

• SMPLIP-Score provides a direct, interpretable, and effective method for predicting binding affinities.
• The model's simplified featurization facilitates understanding and application in drug discovery.
• This approach offers a promising alternative to complex deep learning methods for binding affinity prediction.