LMPred: predicting antimicrobial peptides using pre-trained language models and deep learning.

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Antimicrobial peptides (AMPs) show therapeutic potential in cancer and hypertension and as alternatives to antibiotics due to rising bacterial resistance.
• Experimental identification of AMPs is time-consuming and expensive, necessitating efficient computational screening methods.

Purpose of the Study:

• To develop and validate a novel in silico model for predicting antimicrobial peptides.
• To leverage pre-trained language models for enhanced feature representation of peptide sequences.

Main Methods:

• Utilized pre-trained language models to generate contextualized embeddings for amino acids in peptide sequences.
• Trained a convolutional neural network classifier using these embeddings for AMP prediction.
• Validated the model on a standard dataset and a new, larger independent dataset.

Main Results:

• Achieved predictive accuracies of 93.33% on a previously used dataset.
• Attained 88.26% accuracy on a larger, independent dataset.
• The proposed model outperformed existing state-of-the-art classification models for AMP prediction.

Conclusions:

• The novel approach using language model embeddings and CNNs provides a highly accurate and efficient method for antimicrobial peptide prediction.
• This computational strategy can accelerate the discovery and development of new therapeutic peptides.
• The developed model and code are publicly available to facilitate further research.