PSO Based Neuro-fuzzy Model for Secondary Structure Prediction of Protein

Proteins may be defined as one of the most prominent structural and functional units of life. The structure of a protein is as diverse as the function it sustains. Knowledge about the structural folding of protein serves to ascertain its function. Experimental determination of protein function via its structure is a tedious and gradual process. Protein structure prediction thus; seeks to cultivate adequate ways aimed at providing plausible models for proteins whose structures remain unexplored. Hence, there is a dire need of computational tools to predict, evaluate and visualize the structures of unknown proteins from their amino acid sequences. However, the existing computational tools suffer from various drawbacks that affect their performance substantially like—low prediction accuracy, inefficient modeling of sequence–structure relationship, local methods lacking global exploration, and inability to suffice to the dynamic and exponentially growing data. Accordingly, swarm Intelligence based particle swarm optimization in combination with fuzzy sets has been introduced to propose a neural network based model for secondary structure prediction of protein. The data from six standard datasets namely- RS126, EVA6, CB396, CB513, Manesh and PSS504 has been utilized for the training and testing of the neural network. The model is evaluated using performance measures like- sensitivity, fallout, false discovery rate, miss-rate, specificity, false omission rate, precision, negative predictive value, Q3 accuracy and Matthews correlation coefficient. Sensitivity analysis and 10, 20, 30 and 40 fold cross validation has been performed for further verification of results. The proposed model resolves most of the issues addressed above, thus achieving an average Q3 accuracy of above 90% which is better than the existing tools for secondary structure prediction.