Abstract: Using disordered and ordered proteins collected through literature searches and database scanning, we assembled a set of 24-residue segments centered on order/disorder boundaries and a larger set of non-boundary segments consisting of either order or disorder. We analyzed position specific amino acid compositions around order/disorder boundaries and found that more than thirty compositions differ significantly (p <0.05) between boundary
and non-boundary data. From this analysis, we constructed several logistic regression predictors of order/disorders boundaries using slightly different data modeling approaches. Exact boundary prediction accuracies were estimated to be in the range from 74% to 80% for the different predictors.