GPT2-ICC: A data-driven approach for accurate ion channel identification using pre-trained large language models

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Experimental and computational methods struggle with accurate and efficient ion channel classification in large protein datasets.
• Identifying ion channels is crucial for understanding cellular functions and disease mechanisms.

Purpose of the Study:

• To develop a novel deep learning algorithm for accurate and efficient ion channel classification.
• To address the challenge of imbalanced protein sequence data in ion channel identification.

Main Methods:

• Developed GPT2 Ion Channel Classifier (GPT2-ICC), a deep learning algorithm integrating representation learning with a large language model (LLM).
• Trained and tested GPT2-ICC on a dataset with a significant imbalance of ion channel versus non-ion-channel proteins.
• Evaluated GPT2-ICC's generalization ability by predicting ion channels from the unannotated human proteome.

Main Results:

• GPT2-ICC demonstrated high accuracy in distinguishing ion channels from a dataset with 239 times more non-ion-channel proteins.
• The algorithm successfully predicted several potential ion channels from the unannotated human proteome.
• The study highlighted the effectiveness of combining representation learning with LLMs for analyzing imbalanced biological sequence data.

Conclusions:

• GPT2-ICC represents a significant advancement in artificial-intelligence-driven ion channel research.
• The developed algorithm provides a valuable computational tool for discovering uncharacterized ion channels.
• This approach demonstrates the adaptability of LLMs in addressing complex biological data challenges.