AMPDeep: hemolytic activity prediction of antimicrobial peptides using transfer learning

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Deep learning excels in sequence classification but requires extensive data.
• Predicting hemolytic activity of antimicrobial peptides is challenging due to limited available data.
• Existing methods struggle with small datasets for peptide classification.

Purpose of the Study:

• To develop a deep learning model for accurate hemolytic activity prediction of antimicrobial peptides.
• To address the challenge of limited data in peptide classification tasks.
• To leverage transfer learning for improved performance in antimicrobial peptide analysis.

Main Methods:

• Implemented the AMPDeep pipeline using three distinct datasets for hemolysis activity prediction.
• Utilized a transformer-based model pre-trained on large protein and peptide databases.
• Fine-tuned the model on an aggregated dataset of labeled peptides for supervised prediction.
• Employed transfer learning, hyper-parameter optimization, and selective fine-tuning.

Main Results:

• AMPDeep demonstrated superior performance across three hemolysis activity prediction datasets.
• The model outperformed previous methods relying on physicochemical features or sequence-based deep learning.
• A combined dataset was introduced to mitigate sequence similarity issues in hemolysis prediction.
• AMPDeep successfully leveraged patterns from extensive unlabeled protein and peptide data.

Conclusions:

• Transfer learning effectively overcomes data limitations in predicting antimicrobial peptide hemolysis.
• AMPDeep achieves state-of-the-art results for hemolysis activity classification using only peptide sequences.
• This study facilitates the practical adoption of large sequence-based models for peptide analysis.
• The findings support the use of deep learning for critical peptide classification tasks.