Enhancing Antimicrobial Peptide Function Prediction via Knowledge Transfer on Protein Language Models.

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Antibiotic resistance is a growing global health threat, necessitating the search for alternative antimicrobial agents.
• Traditional experimental methods for identifying antimicrobial peptides (AMPs) are time-consuming and costly.
• Machine learning and deep learning, particularly pre-trained protein language models (pLMs), show promise in accelerating AMP discovery.

Purpose of the Study:

• To develop an efficient computational method for predicting antimicrobial peptides (AMPs) and their specific properties.
• To leverage transfer learning and fine-tuning of pLMs for enhanced AMP prediction accuracy.
• To provide a robust tool that overcomes the limitations of experimental AMP identification.

Main Methods:

• Developed KT-AMPpred, a novel prediction method based on pre-trained protein language models (pLMs).
• Employed transfer learning to integrate knowledge from existing AMP classification tasks.
• Utilized fine-tuning of pLMs to optimize performance for predicting specific antimicrobial properties.

Main Results:

• KT-AMPpred demonstrated superior performance compared to current leading methods on benchmark datasets.
• Visual analysis confirmed the method's robust feature extraction capabilities.
• The model effectively predicts both the presence of AMPs and their distinct antimicrobial characteristics.

Conclusions:

• KT-AMPpred is an effective and efficient tool for antimicrobial peptide prediction.
• The integration of transfer learning and pLM fine-tuning significantly improves prediction accuracy.
• This approach accelerates the discovery of novel AMPs to combat antibiotic resistance.