Scholarlink

Background <p>Antimicrobial peptides (AMPs) are gaining attention as a promising alternative due to their unique ability to disrupt microbial cell membranes and other mechanisms that reduce the likelihood of resistance development. However, traditional experimental methods for identifying and validating AMPs are time-consuming, resource-intensive, and often impractical for screening large peptide libraries. To address these challenges, computational approaches have been developed to automate AMP prediction and characterization, with machine learning (ML) and deep learning (DL) leading thiseffort.</p> Objective <p>The aim of our study was to provide a comprehensive review of existing models for AMP classification and generation, highlighting the strengths of such approaches while emphasizing the weaknesses that hinder the advancement of machine learning in this field.</p> Methods <p>In our study, we generated random 20-amino acid peptides and tested them across multiple trained AMP prediction models.</p> Results <p>Machine learning, particularly with deep neural networks, has become indispensable in AMP research, owing to its capacity to process complex biological data and uncover patterns that can predict the bioactivity of new molecules. Nonetheless, there are inherent limitations. Data scarcity remains a primary obstacle, as high-quality, experimentally validated AMP datasets are still limited. Another critical issue is the prevalence of false positives in ML-based AMP prediction. Surprisingly, a substantial fraction of these random sequences was classified as AMPs (0,5–100%), highlighting potential overfitting and limitations in model generalizability.</p> Conclusions <p>Such false positives underscore the need for improved model robustness, enhanced training datasets, and the incorporation of structural and functional information to refine predictions. Moreover, computational predictions must ultimately be validated through experimental approaches to confirm their antimicrobial activity and therapeutic potential.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

How Machine Learning Helps in Combating Antimicrobial Resistance: A Review of AMP Analysis and Generation Methods

  • 期刊论文

  • 发表日期：

    2025年4月15日

查看全文
  • Khalit Musin,
  • Elvina Asyanova

摘要

Background

Antimicrobial peptides (AMPs) are gaining attention as a promising alternative due to their unique ability to disrupt microbial cell membranes and other mechanisms that reduce the likelihood of resistance development. However, traditional experimental methods for identifying and validating AMPs are time-consuming, resource-intensive, and often impractical for screening large peptide libraries. To address these challenges, computational approaches have been developed to automate AMP prediction and characterization, with machine learning (ML) and deep learning (DL) leading thiseffort.

Objective

The aim of our study was to provide a comprehensive review of existing models for AMP classification and generation, highlighting the strengths of such approaches while emphasizing the weaknesses that hinder the advancement of machine learning in this field.

Methods

In our study, we generated random 20-amino acid peptides and tested them across multiple trained AMP prediction models.

Results

Machine learning, particularly with deep neural networks, has become indispensable in AMP research, owing to its capacity to process complex biological data and uncover patterns that can predict the bioactivity of new molecules. Nonetheless, there are inherent limitations. Data scarcity remains a primary obstacle, as high-quality, experimentally validated AMP datasets are still limited. Another critical issue is the prevalence of false positives in ML-based AMP prediction. Surprisingly, a substantial fraction of these random sequences was classified as AMPs (0,5–100%), highlighting potential overfitting and limitations in model generalizability.

Conclusions

Such false positives underscore the need for improved model robustness, enhanced training datasets, and the incorporation of structural and functional information to refine predictions. Moreover, computational predictions must ultimately be validated through experimental approaches to confirm their antimicrobial activity and therapeutic potential.