Comparative analysis of machine learning based classification for abiotic stress proteins

Abstract

For thousands of years, cereals which include rice, wheat, maize, sorghum and millets etc. have been playing major role in human civilization. These are the principal components of human diet and important staples for daily survival of billions of people globally. The cereal crops belong to poaceae family and rich in vitamins, minerals and fiber. They are reported to reduce the coronary heart disease and other serious diseases. These crops are adversely affected by biotic and abiotic stresses like cold, drought, heat and salinity. With the advent of modern NGS technologies, the plethora of molecular data leads to infer many unexplored facts of the cereal crops using in-silico approach. In the present work, computational techniques were applied to study thoroughly the classification of abiotic stresses (cold, drought, heat and salinity) responsive genes in cereals. The datasets of four stress responsive genes in poaceae family was retrieved from public domain. The machine learning based methodologies namely, Random forest, Support Vector Machines and Deep Learning-Long Short-Term Memory (DL-LSTM) were applied. A comparative analysis was carried out for classification of the retrieved data with k-fold cross validation applying the machine learning techniques at different parameters. It was observed that for all the four sets of data, accuracy was maximum, i.e. 95.11%, 76.88%, 94.31% and 82.04% for cold, drought, heat and salinity, respectively using DL-LSTM. Comparison of the methodologies obviates the outperformance of deep leaning. Such approach of computational studies will help researchers to study the complex biological problems of gene classification more efficiently.