Forman persistent Ricci curvature (FPRC)-based machine learning models for protein-ligand binding affinity prediction

Area of Science:

• Computational Chemistry
• Artificial Intelligence in Drug Discovery
• Topological Data Analysis

Background:

• Artificial intelligence (AI) is increasingly used across the drug design pipeline, from target identification to clinical trials.
• A key challenge in AI-driven drug design is molecular featurization, requiring effective molecular descriptors or fingerprints.
• Efficient and transferable molecular descriptors are crucial for the performance of AI drug design models.

Purpose of the Study:

• To propose a novel molecular featurization and feature engineering method using Forman persistent Ricci curvature (FPRC).
• To develop and evaluate AI models incorporating FPRC-based molecular descriptors for drug design tasks.
• To demonstrate the superiority of FPRC-based descriptors compared to traditional methods.

Main Methods:

• Molecular structures and interactions are represented as simplicial complexes, a higher-dimensional generalization of graphs.
• A filtration process generates a multiscale representation of nested simplicial complexes.
• Forman Ricci curvatures (FRCs) are computed across scales, with their persistence defining FPRC. FPRC-based attributes serve as molecular descriptors, integrated with gradient boosting tree (GBT) models.

Main Results:

• The proposed FPRC-based molecular descriptors were integrated with gradient boosting tree (GBT) models.
• Models were extensively trained and tested on PDBbind-2007, PDBbind-2013, and PDBbind-2016 datasets.
• The FPRC-GBT models demonstrated superior performance compared to machine learning models utilizing traditional molecular descriptors.

Conclusions:

• Forman persistent Ricci curvature (FPRC) offers a novel and effective approach for molecular featurization in AI drug design.
• FPRC-based molecular descriptors enhance the predictive power of machine learning models.
• This method represents a significant advancement in developing efficient and transferable descriptors for AI-based drug discovery.