Protein Secondary Structure Prediction using Pattern Recognition Neural Network

Proteins are key biological molecules with diverse functions. With newer technologies producing more data (genomics, proteomics) than can be annotated manually, in silico methods of predicting their structure and thereafter their function has been christened the Holy Grail of structural bioinformatics. Successful secondary structure prediction provides a starting point for direct tertiary structure modeling; in addition it improves sequence analysis and sequence-structure binding for structure and function determination. Using machine learning and data mining process, we developed a pattern recognition technique based on statistical for predicting protein secondary structure from the component amino acid sequence. By applying this technique, a performance score of Q8=72.3% was achieved. This compares well with other established techniques, such as NN-I and GOR IV which achieved Q3 scores of 64.05% and 63.19% respectively when predictions are made on single sequence alone.