Comprehensive assessment of machine learning-based methods for predicting antimicrobial peptides.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Antimicrobial peptides (AMPs) are crucial for biological functions and are gaining attention due to rising antimicrobial resistance.
• Experimental identification of AMPs is challenging, leading to the development of over 30 computational prediction methods.
• These computational methods vary significantly in datasets, algorithms, and evaluation strategies.

Purpose of the Study:

• To comprehensively survey and compare existing computational methods for AMP identification.
• To evaluate the predictive performance of various AMP prediction tools using independent and curated datasets.
• To benchmark traditional machine learning algorithms for AMP prediction.

Main Methods:

• A comprehensive survey of over 30 computational AMP identification methods.
• Performance evaluation of surveyed tools on an independent test set (1536 AMPs, 1536 non-AMPs).
• Construction of six validation datasets from common AMP databases for comparative analysis.
• 5-fold cross-validation to benchmark machine learning algorithms on a standardized dataset.

Main Results:

• amPEPpy demonstrated superior predictive performance compared to other evaluated methods.
• Random forest, support vector machine, and eXtreme Gradient Boosting showed strong performance in cross-validation.
• Performance variations were observed across different datasets, highlighting the importance of data standardization.

Conclusions:

• amPEPpy is a highly effective tool for antimicrobial peptide prediction.
• Traditional machine learning algorithms like random forest and support vector machine are robust choices for AMP prediction.
• Standardization of datasets is crucial for accurate benchmarking and comparison of computational AMP prediction tools.