Prediction Models for Identifying Ion Channel-Modulating Peptides via Knowledge Transfer Approaches.

Area of Science:

• Computational biology and bioinformatics
• Drug discovery and development
• Molecular pharmacology

Background:

• Ion channels are crucial drug targets for various diseases.
• Experimental screening of ion channel-modulating peptides (IMPs) is costly and time-consuming.
• In-silico methods offer an efficient alternative for IMP screening.

Purpose of the Study:

• To develop PrIMP, a prediction model for screening IMPs targeting sodium, potassium, calcium ion channels, and nicotinic acetylcholine receptors (nAChRs).
• To address data scarcity for IMPs using knowledge transfer techniques.
• To enhance prediction accuracy for IMPs.

Main Methods:

• Utilized multi-task learning (MTL) for simultaneous model training and improved generalization.
• Employed transfer learning (TL) by leveraging pre-trained weights from a larger antimicrobial peptide dataset.
• Integrated deep learning with traditional machine learning for a hybrid approach.

Main Results:

• PrIMP achieved high F1 scores: 0.924 (sodium), 0.937 (potassium), 0.898 (calcium), and 0.931 (nAChRs).
• Both MTL and TL significantly improved prediction performance.
• The hybrid deep learning and machine learning approach further enhanced model accuracy.

Conclusions:

• PrIMP provides an effective in-silico screening tool for IMPs.
• Knowledge transfer methods (MTL and TL) are valuable for overcoming data limitations in peptide prediction.
• The developed model and dataset are publicly available to facilitate further research.