Persistent spectral-based machine learning (PerSpect ML) for protein-ligand binding affinity prediction.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery

Background:

• Molecular descriptors are crucial for QSAR and machine learning in various scientific analyses.
• Existing spectral models lack a multi-scale approach for feature engineering.

Purpose of the Study:

• To introduce persistent spectral-based machine learning (PerSpect ML) models for enhanced drug design.
• To develop a novel feature engineering framework using spectral attributes at multiple scales.

Main Methods:

• A filtration process was introduced to generate spectral models at various scales.
• PerSpect attributes were defined as functions of spectral variables over filtration values.
• Developed PerSpect ML models by combining PerSpect attributes with machine learning for protein-ligand binding affinity prediction.

Main Results:

• PerSpect ML models achieved superior performance on PDBbind-2007, PDBbind-2013, and PDBbind-2016 databases.
• The proposed PerSpect theory offers a powerful feature engineering framework.
• Results indicate performance improvements over all existing models in protein-ligand binding affinity prediction.

Conclusions:

• PerSpect ML models represent a significant advancement in drug design and molecular data analysis.
• The PerSpect theory provides a robust framework for feature engineering in machine learning.
• These models show great potential for improving learning model performance in molecular applications.