Talk by: Kang Peng, IST Center, Temple University

Title: Combining predictors for short and long protein disorder

Abstract: During the past few years we have been focused on improving protein disorder predictors for disordered regions longer than 30 residues. However, these predictors performed considerably worse on disordered regions shorter than 30 residues as observed two years ago during our participation in the CASP5 experiment. To address this problem, we developed a two-level predictor which at the first level consisted of two specialized predictors optimized for long disordered regions (>30 residues) and short disordered regions (<30 residues). At the second level, a meta-predictor was built to determine which of the two first-level predictors should be used at a given sequence position. Ideally, the two specialized predictors should receive weights of 1/0 in long disordered regions, 0/1 in short disordered regions, and 0.5/0.5 in ordered regions. As the results in the latest CASP6 experiment showed, this new predictor achieved significantly better prediction accuracy on short disordered regions, while its performance on long disordered regions was comparable to that of our previously developed long disorder predictors. The reported results were obtained through a collaboration with A.K. Dunker, P. Radivojac, S. Vucetic and Z. Obradovic funded by NIH-R01-LM007688-01A1 research grants.