Can machine learning 'transform' peptides/peptidomimetics into small molecules? A case study with ghrelin receptor...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Pharmacology

Background:

• Peptide-based drugs face challenges with bioavailability and metabolic stability.
• Transforming peptides into small molecules is a key strategy in drug development.
• The ghrelin receptor (GR) is a therapeutic target implicated in various diseases.

Purpose of the Study:

• To develop machine learning (ML) models for identifying bioactive peptide features.
• To assess ML's ability to discriminate between binding and non-binding small molecules.
• To predict small molecule binders for the ghrelin receptor using ML models trained on peptide data.

Main Methods:

• Constructed ML models using random forest, support vector machine, and extreme gradient boosting algorithms.
• Utilized a curated dataset of peptide/peptidomimetic and small molecule GR ligands.
• Trained and validated models using both peptide-exclusive and diverse peptide/small molecule datasets.

Main Results:

• ML models trained solely on peptides showed limited predictive power for small molecules.
• Models trained on diverse datasets (peptides and small molecules) demonstrated high accuracy.
• Diversified models successfully differentiated binding from non-binding small molecules, including newly synthesized compounds.
• Identified critical structural features for binding activity consistent with experimental data.

Conclusions:

• Diverse datasets significantly improve ML model performance for predicting small molecule drug binders.
• ML models can effectively identify key structural features driving ligand-receptor interactions.
• This approach offers a promising strategy for small molecule drug discovery targeting receptors like GR.