Hierarchical machine learning model predicts antimicrobial peptide activity against Staphylococcus aureus

Frontiers in Molecular Biosciences

Molecular Biosciences

Hosein Khabaz¹  Amir Homayoun Keihan¹  Mehdi Rahimi-Nasrabadi² 

[1] Molecular Biology Research Center, Systems Biology and Poisonings Institute, Baqiyatallah University of Medical Sciences, Tehran, Iran;Molecular Biology Research Center, Systems Biology and Poisonings Institute, Baqiyatallah University of Medical Sciences, Tehran, Iran;Faculty of Pharmacy, Baqiyatallah University of Medical Sciences, Tehran, Iran;

关键词: Staphylococcus aureus;    antimicrobial peptides;    machine learning;    antimicrobial activity;    classification model;   

DOI  :  10.3389/fmolb.2023.1238509

 received in 2023-06-11, accepted in 2023-08-31,  发布年份 2023

来源: Frontiers

【 摘 要 】

Introduction: Staphylococcus aureus is a dangerous pathogen which causes a vast selection of infections. Antimicrobial peptides have been demonstrated as a new hope for developing antibiotic agents against multi-drug-resistant bacteria such as S. aureus. Yet, most studies on developing classification tools for antimicrobial peptide activities do not focus on any specific species, and therefore, their applications are limited.Methods: Here, by using an up-to-date dataset, we have developed a hierarchical machine learning model for classifying peptides with antimicrobial activity against S. aureus. The first-level model classifies peptides into AMPs and non-AMPs. The second-level model classifies AMPs into those active against S. aureus and those not active against this species.Results: Results from both classifiers demonstrate the effectiveness of the hierarchical approach. A comprehensive set of physicochemical and linguistic-based features has been used, and after feature selection steps, only some physicochemical properties were selected. The final model showed the F1-score of 0.80, recall of 0.86, balanced accuracy of 0.80, and specificity of 0.73 on the test set.Discussion: The susceptibility to a single AMP is highly varied among different target species. Therefore, it cannot be concluded that AMP candidates suggested by AMP/non-AMP classifiers are able to show suitable activity against a specific species. Here, we addressed this issue by creating a hierarchical machine learning model which can be used in practical applications for extracting potential antimicrobial peptides against S. aureus from peptide libraries.

【 授权许可】

Unknown   
Copyright © 2023 Khabaz, Rahimi-Nasrabadi and Keihan.