Ensembling machine learning models to boost molecular affinity prediction.

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Predicting molecular binding affinity is crucial for drug discovery.
• Accurate prediction can accelerate the identification of potential drug candidates.
• Existing methods may rely on structural data, introducing potential biases.

Purpose of the Study:

• To develop and evaluate a machine learning pipeline for predicting molecular binding affinity.
• To integrate diverse machine learning models for enhanced prediction accuracy.
• To provide a structure-agnostic approach, avoiding bias from known molecular conformations.

Main Methods:

• Six machine learning algorithms were employed: Support Vector Machine, Random Forest, CatBoost, feed-forward neural network, graph neural network, and Bidirectional Encoder Representations from Transformers.
• Models were trained using ligand features based on physico-chemical properties and textual representations.
• A two-stage ensemble pipeline was created, first for classification (binding or not) and then for regression (binding affinity K_i).

Main Results:

• The study successfully integrated multiple machine learning approaches to predict molecular binding affinity.
• The ensemble pipeline demonstrated robust performance in both classification and regression tasks.
• The models generalize to compounds with unknown conformations, as they do not rely on atomic spatial coordinates.

Conclusions:

• The proposed machine learning pipeline offers a powerful and flexible tool for predicting molecular binding affinity.
• This approach can be repurposed for various receptors beyond human thrombin.
• The structure-agnostic nature enhances its applicability in drug discovery and development.