TooT-BERT-C: A study on discriminating ion channels from membrane proteins based on the primary sequence's contextual representation from BERT models

While ion channels play a critical role in a variety of physiological processes and are a frequent therapeutic target, their function that contributes to disease remains unknown. Computational techniques have emerged as crucial and indispensable tools for understanding ion channels and their function in recent years. This is because their mechanism of action is complex, and a static representation of an ion channel is frequently insufficient to comprehend the underlying process. This article introduces TooT-BERT-C, a technique that utilizes the BERT contextual representation to assess and discriminate ion channels from membrane proteins via a Logistic Regression classifier. Additionally, we compare two alternative BERT models' frozen and fine-tuned representations, namely ProtBERT-BFD and MembraneBERT. When compared to leading deep learning prediction algorithms, TooT-BERT-C has the highest accuracy of 98.24 percent and MCC of 0.85.