Protein Secondary Structure Detection without Alignment by Recurrent Neural Network with LSTM

Studying the biological functionality of spices, predicting protein secondary structure from amino acid sequences is a complicated task as it has become a must to find similarities or dissimilarities between protein bodies. However, research has been conducted to predict protein secondary structure, yet the accuracy rate is not satisfactory for large datasets. Here, we propose a model to detect protein 2D structure without alignment by recurrent neural network with bi-directional LSTM named as Protein 2D Prediction Algorithm (P2DPA). In this modern era, Protein Data Bank (PDB) is growing rapidly. Therefore, P2DPA is not only a faster model but also it has a better accuracy on large datasets. Some previous models like X-ray crystallography (XRC), Nuclear Magnetic Resonance (NMR) are costly and have less accuracy with respect to P2DPA. A machine learning approach has been introduced in P2DPA which is faster and more accurate for complex datasets as well as it can achieve faster computational time with less memory resources. We have tested our model on DSSP, CB513, CASP12, TS115, and RCSB datasets and achieved a satisfactory accuracy result with respect to time and space complexity.