Protein secondary structure prediction using BLAST and relaxed threshold rule induction from coverings

Protein structure prediction has always been an important research area in bioinformatics and biochemistry. Despite the recent breakthrough of combining multiple sequence alignment information and artificial intelligence algorithms to predict protein secondary structure, the Q3 accuracy of various computational prediction methods rarely has exceeded 75%; this status has changed little since 2003 when Rost stated that "the currently best methods reach a level around 77% three-state per-residue accuracy.";The application of artificial neural network methods to this problem is revolutionary in the sense that those techniques employ the homologues of proteins for training and prediction. In this dissertation, a different approach, RT-RICO (Relaxed Threshold Rule Induction from Coverings), is presented that instead uses association rule mining. This approach still makes use of the fundamental principle that structure is more conserved than sequence. However, rules between each known secondary structure element and its "neighboring" amino acid residues are established to perform the predictions. This dissertation consists of five research articles that discuss different prediction techniques and detailed rule-generation algorithms. The most recent prediction approach, BLAST-RT-RICO, achieved a Q3 accuracy score of 89.93% on the standard test dataset RS126 and a Q3 score of 87.71% on the standard test dataset CB396, an improvement over comparable computational methods.;Herein one research article also discusses the results of examining those RT-RICO rules using an existing association rule visualization tool, modified to account for the non-Boolean characterization of protein secondary structure.