Deep Learning for Predicting Residue ZCoordinates in _-Helical Transmembrane Proteins (TM-ZC)

Abstract

Z-coordinate, defined as the residue's distance from the center of the biological membrane, is a crucial
structural property of -helical transmembrane proteins (-TMPs). Neither experimentally solved nor
computationally anticipated -TMP structures can z-coordinate prediction allows us to partially describe -TMP
structures based on their sequences, which helps with function annotation and drug target finding, and so
meets the needs of the relevant study fields. To enhance prediction accuracy and provide a useful tool, we
suggested a deep learning-based predictor (TM-ZC) for the z-coordinate of residues in -TMPs. TM-ZC trained a
convolutional neural network (CNN) regression model using the one-hot code and the PSSM as input features.
The experimental findings showed that TM-ZC was an effective predictor that is both easy to use and quick to
run, with respectable results: an average error of 1.958, a percent of prediction error within 3 of 77.461%, and
a correlation coefcient (CC) of 0.922. We went on to explore how the TM-ZC predicted z-coordinate may be
helpful, and we discovered that it has a high degree of consistency with topological structure and improves the
prediction of surface accessibility.