PepExplainer: An explainable deep learning model for selection-based macrocyclic peptide bioactivity prediction and...

Area of Science:

• Drug discovery and development
• Computational chemistry
• Artificial intelligence in medicine

Background:

• Macrocyclic peptides are promising drug candidates but costly to test.
• Current AI models for bioactivity prediction face data limitations and interpretability issues.

Purpose of the Study:

• To develop an explainable AI model, PepExplainer, for accurate macrocyclic peptide bioactivity prediction.
• To address challenges of limited data and model interpretability in AI-driven drug discovery.

Main Methods:

• Developed PepExplainer, a graph neural network using substructure mask explanation (SME).
• Translated macrocyclic peptides into atomic-level molecular graphs, handling complex structures and non-canonical amino acids.
• Utilized peptide enrichment and bioactivity data with transfer learning for improved predictions.

Main Results:

• PepExplainer demonstrated effectiveness in predicting macrocyclic peptide bioactivity, validated on newly synthesized peptides.
• The model successfully optimized a macrocyclic peptide's IC50 from 15 nM to 5.6 nM.
• PepExplainer identified key molecular patterns contributing to bioactivity.

Conclusions:

• PepExplainer enhances AI-driven drug discovery by providing interpretable predictions for macrocyclic peptides.
• The model can accelerate the identification and optimization of novel peptide-based therapeutics.
• Explainable AI offers a powerful approach to overcome limitations in traditional drug development pipelines.