LSTM-GRU Based Deep Learning Model with Word2Vec for Transcription Factors in Primates

Abstract

The study of the structures of proteins and the relationships of amino acids remains a challenging problem in biology. Although some bioinformatics-based studies provide partial solutions, some major problems remain. At the beginning of these problems are the logic of the sequence of amino acids and the diversity of proteins. Although these variations are biologically detectable, these experiments are costly and time-consuming. Considering that there are many unclassified sequences in the world, it is inevitable that a faster solution must be found. For this reason, we propose a deep learning model to classify transcription factor proteins of primates. Our model has a hybrid structure that uses Recurrent Neural Network (RNN) based Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks with Word2Vec preprocessing step. Our model has 97.96% test accuracy, 97.55% precision, 95.26% recall, 96.22% f1-score. Our model was also tested with 5-fold cross-validation and reached 97.42% result. In the prepared model, LSTM was used in layers with fewer units, and GRU was used in layers with more units, and it was aimed to make the model a model that can be trained and run as quickly as possible. With the added dropout layers, the overfitting problem of the model is prevented.