A Method for Predicting Hemolytic Potency of Chemically Modified Peptides From Its Structure.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning in drug discovery

Background:

• Hemolytic peptides can be therapeutic or toxic.
• Predicting hemolytic potency is crucial for peptide drug development.
• Computational methods can accelerate the identification of safe and effective peptides.

Purpose of the Study:

• To develop accurate machine learning models for predicting the hemolytic potency of chemically modified peptides.
• To identify key peptide features that influence hemolytic activity.
• To provide a user-friendly tool for researchers studying hemolytic peptides.

Main Methods:

• Dataset compilation of 583 modified hemolytic and non-hemolytic peptides.
• Feature extraction including 2D/3D descriptors, fingerprints, and atom/diatom compositions.
• Development and evaluation of machine learning classification models, including Random Forest.

Main Results:

• The Random Forest model using peptide fingerprints achieved the highest accuracy (78.33%) and AUC (0.86) on the training dataset.
• Validation dataset performance was comparable, with 78.29% accuracy and 0.85 AUC.
• The study identified specific peptide features that are predictive of hemolytic activity.

Conclusions:

• Machine learning models can effectively predict hemolytic peptide potency.
• Peptide fingerprints are valuable features for this prediction task.
• The HemoPImod web server provides a valuable resource for the scientific community to predict hemolytic activity.