Machine learning-enabled predictive modeling to precisely identify the antimicrobial peptides.

Area of Science:

• Computational Biology and Cheminformatics
• Drug Discovery and Development

Background:

• Antimicrobial peptides (AMPs) show broad-spectrum activity against microbes.
• Multi-drug resistant infections pose a significant global health threat.
• In silico prediction of AMPs can accelerate the discovery of new therapeutics.

Purpose of the Study:

• To develop and validate machine learning models for predicting antimicrobial peptides (AMPs).
• To identify key features contributing to the antimicrobial activity of AMPs.
• To provide a tool for the rational design and prediction of novel AMPs.

Main Methods:

• Utilized a large dataset of 2638 AMPs and 3700 non-AMPs.
• Trained and evaluated various machine learning classifiers: Random Forest (RF), k-Nearest Neighbors (k-NN), Support Vector Machine (SVM), Decision Tree (DT), Naive Bayes (NB), Quadratic Discriminant Analysis (QDA), and ensemble learning.
• Performed internal and external cross-validation to assess model performance.

Main Results:

• The Random Forest (RF) classifier demonstrated superior performance, achieving high accuracy in both internal (91.40%) and external (89.43%) validations.
• Key predictive features identified include ChargeD2001, PAAC12 (pseudo amino acid composition), and polarity T13.
• The RF model accurately classified known AMPs and non-AMPs in an additional validation set.

Conclusions:

• The developed RF-based classification model is effective for in silico prediction of antimicrobial peptides.
• Identified features provide insights into the structural determinants of AMP activity.
• This model can serve as a valuable tool for accelerating the discovery and design of novel AMPs to combat antimicrobial resistance.