PREDICTION OF VESICULAR TRANSPORT PROTEINS FROM SEQUENCE DATA WITH GRU-BASED DEEP LEARNING MODELS

Abstract

Prediction of protein function is a very active area in the field with many computational biology researchers working on the problem. It still presents a major challenge to apply deep neural networks in improving the prediction of protein function. In this paper, we present a new method to deal with vesicular transport proteins which are biologically important and use a gated recurrent unit and position-specific scoring matrix profiles. Moreover, despite the difficulty to explain its function, our model (ATCPPP) got accuracies of PseAAC 93.75% and DPC 86.47% in cross-validation datasets on two metrics. We handle the problem of dataset imbalance by implementing class weights in the deep learning model. The results obtained sensitivity, specificity MCC and AUC of 93.75%, 86.47% 91.55 and 90.41%, respectively. We show state of the art results using our methods on the same dataset better than all previous approaches. Here, in our proposed study we suggested to include the identification of other classes of proteins those involved in vesicle transport. In addition, our success may stimulate the use of gated recurrent unit architecture in protein function prediction