A PLM-based method for predicting protein ion channel modulators for drug discovery and safety evaluation

Area of Science:

• Biochemistry and Molecular Biology
• Pharmacology
• Computational Biology

Background:

• Ion channels are critical for physiological processes like neuronal communication and muscle contraction.
• Modulation of ion channels offers therapeutic potential but carries risks of adverse effects.
• Accurate prediction of ion channel modulators is essential for drug development and safety.

Purpose of the Study:

• To develop and validate a predictive model for identifying protein ion channel modulators.
• To analyze sequence composition and identify key residues in ion channel modulating proteins.
• To create a user-friendly web server for drug discovery and biosafety evaluation.

Main Methods:

• Utilized a large, non-redundant dataset of ion channel modulating proteins for training and testing.
• Explored alignment-based and alignment-free methods, including machine learning, deep learning, and protein language models (PLMs).
• Employed an evolutionary information-based PLM (ESM2-t33), integrating BLAST output for enhanced performance.

Main Results:

• Identified Cys, Gly, and Trp as prevalent residues, and Ala, Glu, Leu, Gln, Val as scarce in modulating proteins, with Cys being discriminative.
• Achieved high performance with an AUROC of 0.97 (0.98 with BLAST integration) using the PLM-based approach.
• The IonNTxPred method demonstrated superior performance over existing approaches on independent datasets.

Conclusions:

• IonNTxPred, a PLM-based tool, effectively predicts protein ion channel modulators.
• The tool supports drug repurposing, discovery of new therapeutic proteins, and biosafety evaluations.
• The IonNTxPred web server is available for facilitating protein-based drug design and neurotoxin screening.