This resource is about prediction of Secondary Structures of Proteins (SSP) or peptides based on structural formula of molecular fragments. The classification sequence-structure-property relationships (SSPR) models were created by MultiPASS software on the data on types of SSP annotated by DSSP method (DSSP 4.0.4) for 3D structures of proteins from PDB with a resolution of 2 angstroms or less. The training set included more 323 000 peptides annotated as SSP with lengths from 4 to 60 amino acids represented as structural formulas. The mean accuracy of prediction for the SSPR models based on 9 level of MNA descriptors (AUC) for 8 types of PSS is 0.902.
Prediction is based on special version of PASS (Prediction of Activity Spectra for Substances) technology – MultiPASS based on Bayesian algorithm and MNA (Multilevel Neighborhoods of Atoms) descriptors. The list of predicted types of PSS with their number and accuracy of prediction is available at description of Training Sets .