A BERT-based approach for identifying anti-inflammatory peptides using sequence information.

Area of Science:

• Computational biology
• Drug discovery
• Immunology

Background:

• Anti-inflammatory peptides (AIPs) offer therapeutic potential for inflammatory diseases.
• Experimental identification of AIPs is costly and challenging.
• Computational approaches are emerging as a promising alternative for AIP discovery.

Purpose of the Study:

• To develop a novel computational method, BertAIP, for predicting anti-inflammatory peptides (AIPs).
• To evaluate BertAIP's performance against existing methods for AIP prediction.
• To enhance the interpretability of the prediction model.

Main Methods:

• Utilized a bidirectional encoder representation from transformers (BERT) model for feature extraction from amino acid sequences.
• Employed a fully connected feed-forward network for AIP classification.
• Trained and evaluated the model using AIP datasets from the Immune Epitope Database.

Main Results:

• BertAIP achieved an accuracy of 0.751 and a Matthews correlation coefficient of 0.451.
• Performance metrics surpassed those of commonly used prediction methods.
• Independent testing confirmed BertAIP's superiority over existing AIP predictors.
• Identified and visualized key amino acids influencing AIP prediction.

Conclusions:

• BertAIP is an effective tool for predicting anti-inflammatory peptides (AIPs) based solely on amino acid sequences.
• The model demonstrates superior performance compared to current predictors.
• BertAIP can facilitate large-scale screening and the identification of novel AIPs for therapeutic research and drug development in inflammatory diseases.