Protein class prediction based on Count Vectorizer and long short term memory

Abstract

Proteins class and function prediction is one of the most significant task in computational bioinformatics. The information about the protein functions and class plays a vital role in understanding biological cells and has a great impact on human life in factors such as personalized medicine. The technical advancement in the areas of biological aspects and understanding of biological processes results in features and characteristics of important Proteins. Prediction of amino acid sequence involves prediction of amino sequence folding and its structures from the primary sequence obtained. In this work, Machine learning prediction algorithms have applied for protein class prediction. This method takes consideration of macromolecules of biological significances. Later the solution focuses on the understanding of different protein family, subsequently classify the protein family type sequence. This is achieved through machine learning algorithms Naive Bayes (NB) and Random forest (RF) algorithms with count vectorized feature and LSTM. These algorithms are used to classify the protein family on its protein sequence. Finally, result shows that LSTM predicts the protein class more accurately than the RF, and NB algorithm. LSTM achieves an accuracy of 96% whereas RF & NB with an accuracy of 91% and 86%.