TCS-TP: Transporter Protein Prediction Based On Multi-Scale Feature Extraction

Transporter proteins play a crucial role in maintaining ionic homeostasis inside and outside the cell, facilitating protein uptake, and enabling cellular communication with the external environment. Transporter proteins (TPs) are also significant targets for medical research and drug development. Accurately predicting novel TPs remains a major challenge in functional genomics. In this study, we propose TP prediction models, designated as TCS-TP, that utilize transformer and convolutional neural networks for multi-scale feature extraction of protein sequences. The transformer incorporates the GLU activation function, whereas the convolutional neural network (CNN) features three parallel subnetworks. Support vector machines are used as a classifier for TP classification. The test results demonstrate that TCS-TP could successfully recognize TPs, with AUROC of 0.89, AUPRC of 0.81, and accuracy of 91.6617%. Upon further comparison, it is determined that TCS-TP outperforms other methods. We hoped that TCS-TP will prove to be a valuable tool for predicting TPs in large-scale genomic projects and contribute to the discovery of new TPs.