Integrated convolution and self-attention for improving peptide toxicity prediction.

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Peptides offer therapeutic potential but face development hurdles due to toxicity.
• Current toxicity testing methods are slow and expensive, hindering clinical application.
• Accurate and rapid computational prediction of peptide toxicity is crucial for drug development.

Purpose of the Study:

• To introduce CAPTP, a novel computational approach for predicting peptide toxicity.
• To leverage convolutional and self-attention mechanisms for enhanced toxicity prediction.
• To facilitate the identification of safe peptide candidates for drug development.

Main Methods:

• Utilized convolutional neural networks and self-attention mechanisms.
• Developed a computational tool named CAPTP for sequence-based toxicity prediction.
• Validated performance using cross-validation and independent test datasets.

Main Results:

• CAPTP achieved a Matthews correlation coefficient of approximately 0.82.
• Performance surpassed existing state-of-the-art peptide toxicity predictors.
• Identified specific sequential patterns in peptide head and central regions critical for toxicity.

Conclusions:

• CAPTP offers a robust and generalizable method for peptide toxicity prediction.
• The tool's insights can guide the rational design of safer peptide therapeutics.
• Freely available source code promotes accessibility and further research.