Related Experiment Video
Updated: Aug 14, 2026
13:49
Semi-automated Biopanning of Bacterial Display Libraries for Peptide Affinity Reagent Discovery and Analysis of Resulting Isolates
Published on: December 6, 2017
An Interpretable Multi-Objective Machine Learning Framework for In Silico Prioritization of Anti-Staphylococcus
Jianguo Xu1,2, Donghua Yang1,3, Qingyong Zheng1
1Evidence-Based Medicine Center, School of Basic Medical Sciences, Lanzhou University, Lanzhou 730000, China.
Diagnostics (Basel, Switzerland)
|August 13, 2026
Summary
This study developed a machine learning pipeline to identify antimicrobial peptides (AMPs) effective against Staphylococcus aureus with reduced host toxicity. The pipeline prioritizes optimized known AMPs, offering computational hypotheses for further development.
Area of Science:
- Computational biology and bioinformatics
- Antimicrobial drug discovery
- Machine learning in pharmacology
Background:
- Staphylococcus aureus infections, including methicillin-resistant strains, are a significant clinical challenge.
- Antimicrobial peptides (AMPs) show promise but require careful selection to balance efficacy and host toxicity.
- Existing machine learning models may overestimate performance due to data homology.
Purpose of the Study:
- To develop and validate an interpretable, multi-objective machine learning pipeline for identifying anti-Staphylococcus aureus AMPs.
- To minimize predicted host toxicity (hemolysis) while maximizing antimicrobial potency.
- To rigorously benchmark model performance using homology-aware cross-validation.
Main Methods:
- Curated datasets of S. aureus-active AMPs and hemolysis records.
- Peptide encoding using 538 physicochemical and compositional features.
- Evaluation of classifiers and regressors under random and homology-aware cross-validation.
- Model interpretation using SHAP and k-mer enrichment.
- Ranking candidates using a multi-objective score.
Main Results:
- Homology-aware validation yielded lower potency prediction performance (AUROC ~0.71) than random splitting.
- Hemolysis prediction remained robust (AUROC 0.90), indicating accuracy is not due to homology leakage.
- Identified key features driving potency (charge, amphipathicity) and hemolysis (hydrophobicity).
- Prioritized 20 candidates, identified as optimized variants of known AMP scaffolds with high sequence identity.
Conclusions:
- An interpretable, honestly benchmarked pipeline was developed to optimize existing anti-S. aureus AMP scaffolds.
- Prioritized peptides are computational hypotheses requiring experimental validation for efficacy and safety.
- The pipeline successfully integrated potency and low hemolysis criteria, but cross-species selectivity was not assessed.
