Alignment-Free Antimicrobial Peptide Predictors: Improving Performance by a Thorough Analysis of the Largest...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Cheminformatics
• Drug Discovery and Development

Background:

• Numerous machine-learning predictors for antimicrobial peptide (AMP) activities exist, but suffer from unrepresentative training data and sequence duplication issues.
• These limitations reduce the confidence in existing models for prospective studies and reliable AMP identification.
• Addressing these drawbacks is crucial for advancing AMP research and therapeutic applications.

Purpose of the Study:

• To develop novel, high-performance machine learning models for predicting AMP activities.
• To create reliable modeling and assessment datasets from the largest experimentally validated, non-redundant peptide dataset.
• To identify general AMPs and their specific functional types (antibacterial, antifungal, antiparasitic, antiviral) with high accuracy.

Main Methods:

• Developed alignment-free quantitative sequence-activity models (AF-QSAMs) using Random Forest on novel, curated datasets.
• Performed applicability domain analysis for the first time in AMP recognition to ensure prediction reliability.
• Benchmarked proposed models against 13 existing literature programs using rigorous validation tests.

Main Results:

• The proposed AF-QSAM models demonstrated superior performance across all modeled endpoints compared to existing literature methods.
• Applicability domain analysis confirmed the reliability of predictions, a novel approach for AMP recognition.
• Benchmarking revealed that most literature methods exhibited weak-to-random predictive agreement, unlike the proposed models.

Conclusions:

• The novel models, built upon improved datasets and methods, significantly enhance the identification of AMPs with antibacterial, antifungal, antiparasitic, and antiviral activities.
• The developed models offer high effectiveness and reliability, overcoming limitations of previous approaches.
• Freely available via the AMPDiscover tool, these models facilitate future research and development of novel antimicrobial therapies.