A Analysis of Artificial Neural Networks use in Predicting Secondary Structures

Knowing the roles of proteins is crucial given the growing significance of proteins in daily life. Understanding protein structure explains how proteins work. The experimental methods used to analyse protein structures are labor-intensive, expensive, and highly skilled. Therefore, it is crucial to design a workable quantitative strategy. This area of study focuses on protein-secondary-structure predictions, that is strongly compared to its tertiary structure, due to the tremendous complexity of directly predicting protein structure (tertiary structure). In this study, a range of features, such as position-specific scoring matrices, hidden Markov model alignment factorization, and physicochemical factors, are studied in order to anticipate the secondary structure. Additionally, it tries to investigate a good mix of traits that could capture various details about the protein secondary structure. The models are then fitted with an artificial neural network, and evaluation measures such as accuracy, specificity, recall, f1-score, and running time are compared across different datasets. Our suggested model outperforms the benchmark models by at least 2.9% when tested on the CB513 dataset and CB6133 dataset, respectively.