Bulbul Ahmed

Cereals are staple crops widely cultivated across the world. These are highly nutritious, rich in vitamins, minerals, carbohydrates, fats, oils, proteins and fibers but are low in essential amino acids such as lysine. Cereal crops belong to poaceae family, having wider applications in production of flour, bread, rice, cakes, corn etc. The other by-products of these crops are beverages and wine. Moreover, consumption of these crops reduces the coronary heart disease, diabetes, colon cancer, diverticular disease etc. India is the third largest cereal producer after China and USA but it has been producing to a great extent which could be achieved to 4.9% increase in production from base year 2020 to 2027. The production of these crops is highly affected by biotic and abiotic stresses which adversely affected crop growth and development, further resulting in crop loss that leads to economic loss. Hence, it is required to understand and study the genes involved in order to minimize the biotic and abiotic stresses. The genes start adapting under stress factors and produce proteins that can tolerate such changes by changing signalling pathways in protein-protein interaction. Finding these proteins are highly expensive, time consuming and required a highly experienced person. In order to reduce cost and time, rapid classification and prediction of such proteins using computation approaches is required. Further, these proteins are complex in nature with high dimensions which are very difficult to study using conventional approaches. This study was oriented towards the application of different machine learning techniques (namely, support vector machine and random forest) and deep learning (long short-term memory) for development of classification models for abiotic stresses (heat, cold, salinity and drought) protein sequences from poaceae family. Also, an activation function, Gaussian Error Linear Unit with Sigmoid function (SiELU) has been developed for deploying in a deep learning model which shows an increased efficiency of the model. Lastly, a web-based tool for prediction of stress associated proteins from poaceae family has been developed implementing the proposed long short-term memory deep learning methodology with developed activation function i.e., SiELU and tuning of other hyper-parameters.