Local Similarity Matrix for Cysteine Disulfide Connectivity Prediction from Protein Sequences

AbstractAccurately predicting three dimensional protein structures from sequences would present us with targets for drugs via molecular dynamics that would treat cancer, viral infections, and neurological diseases. These treatments would have a far reaching impact to our economy, quality of life, and society. The goal of this research was to build a data mining framework to predict cysteine connectivity in proteins from the sequence and oxidation state of cysteines. Accurately predicting the cysteine bonding configuration improves the TM-Score, a quantitative measurement of protein structure prediction accuracy. We provided state of the art Qp and Qc on the PDBCYS and IVD-54 Datasets. Furthermore, we have produced a Local Similarity Matrix that compares favorably to the default PSSMs generated from PSI-Blast in a statistically significant way. Our Qp for SP39, PDBCYS, and IVD-54 were 90.6, 80.6, and 68.5, respectively.