Saptashwa Datta and colleagues at Asia University and National Chung Hsing University introduce AAGP, which leverages 4,305 physicochemical and compositional features and employs Boruta-based ranking and heuristic subset selection to train gradient boosting and extra-trees models, achieving MCCs up to 0.692 in distinguishing anti-aging peptides, thus advancing peptide therapeutic design.
Key points
- Encoded peptide sequences into 4,305 physicochemical and compositional features and selected the top 50 via Boruta ranking and heuristic forward selection.
- Applied Bayesian-optimized gradient boosting (LGBM) and extra-trees classifiers to achieve independent test MCCs of 0.692 and 0.580 with AUCs of 0.963 and 0.808.
- SHAP analysis and residue property correlations reveal distinct reliance on physicochemical features for antimicrobial-based negatives and compositional features for random peptide negatives.
Why it matters: This AI-driven platform streamlines anti-aging peptide discovery, offering a scalable strategy to accelerate development of peptide therapeutics targeting skin and age-related conditions.
Q&A
- What are anti-aging peptides?
- How does AAGP predict peptide activity?
- What are physicochemical features?
- Why use distinct negative datasets?
- What is Matthew's correlation coefficient (MCC)?