A Machine Learning Approach for Predictive Insights into Tight Junction Protein Functions

Abstract

Tight junction proteins are a crucial type of protein in the structure of cell membranes in multicellular organizations, such as the intestinal epithelium, urinary tract epithelium, and other tissues. They hold adjacent cells tightly together, forming a barrier that prevents the passage of fluids and other molecules between cells. Tight junctions maintain a distinct environment inside and outside the cells, preventing uncontrolled exchange of fluids, ions, and molecules between cells. They play a vital role in maintaining the distinctiveness of organs and mucous membranes in the body. Dysfunction of tight junction proteins can lead to various pathological conditions in humans. However, determining the functions of tight junction proteins has traditionally been conducted through experiments, a process that consumes considerable time, effort, and resources. In this study, we present a novel methodology utilizing both traditional machine learning models and advanced deep learning models for the prediction of tight junction protein functions. Our model combines features with the integration of multiple biologically relevant characteristics extracted from input sequence data. Overall, our model yields promising results, with an average accuracy is 90,1% and AUC is 0,953 on an independent test dataset for tight junction protein classification. This demonstrates the reliability of our proposed model for predicting the functions of tight junction proteins. Predictions through this new model will assist researchers in rapidly predicting the functions of tight junctions, saving time, effort, and resources compared to experimental predictions.