Protein Language Models and Machine Learning Facilitate the Identification of Antimicrobial Peptides

Area of Science:

• Biochemistry and Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Peptides are versatile bioactive molecules with broad applications.
• Increasing peptide data necessitates advanced computational methods for sequence-function analysis.
• Accelerating the discovery and design of antimicrobial peptides (AMPs) is crucial for therapeutic development.

Purpose of the Study:

• To introduce AMP-Detector, a sequence-based classification model for predicting peptide biological activity.
• To focus on accelerating the discovery and de novo design of potential antimicrobial peptides (AMPs).
• To develop a robust computational pipeline integrating protein language models and machine learning.

Main Methods:

• Developed a novel sequence-based pipeline for training binary classification models.
• Integrated protein language models with machine learning algorithms.
• Trained 21 models targeting antimicrobial, antiviral, and antibacterial activities.

Main Results:

• Achieved average precision exceeding 83% for peptide activity prediction.
• Outperformed existing methods for AMP detection and showed comparable results for other activities.
• Discovered over 190,000 potential AMPs and designed over 500 novel AMPs using a generative approach.

Conclusions:

• AMP-Detector offers a significant advancement in peptide-based drug discovery.
• The integrative approach combining robust models and generative design aids in de novo peptide design.
• Represents a pivotal tool for therapeutic applications, particularly in antimicrobial peptide development.