SSTmPROT, A POINT mu TATION SENSITIVE TOO l T O CO mb INE RES ulTS AND TO PREDICT THE CONSENS u S SEQ u ENCE , AND SECONDARY STR u CT u RE OF TRANS m E mb RANE PROTEINS

Abstract: The efficiency of predictions of protein secondary structures can be increased by treating this process as a cycle of steps, where each step is an approach to the single natural event in folding process (model and simulation of successive events). The set of simulation steps qualifies reliability of in silico simulation of single steps, allowing to verify the correctness of each step as well as to retain sensitivity in case of a single amino acidic substitution. For this purpose the three-part algorithm (SSTMProt) has been designed. This algorithm combines the results of known methods of prediction of proteins secondary structure. Furthermore, the efficiency of this algorithm has been verified using the models received from the RCSB PDB (the Research Collaboratory for Structural Bioinformatics – Protein Data Base; http://www.rcsb.org). The accuracy of known methods has been compared with the accuracy of designed algorithms. The accuracy has been tested by the comparison of true secondary structure with predicted secondary structure of a given protein. The results of accuracy test has been presented as percentage values of similarity between both secondary structures: predicted structure using known method vs. true structure and predicted structure using designed method vs. true structure. The results demonstrate 20-30% higher accuracy of prediction for designed algorithms then for adequate known methods. The test of sensitivity has been done for proteins of a very conservative and stable structure (subunits of bovine cytochrome c oxidase and bacterial ATP-ase, bovine rhodopsin and human hemoblobin as a globular but alpha-helical protein). The influence of a single amino acid substitution on a resulted secondary structure predicted by SSTMProt algorithms has been examined. The repeatability of elaborated algorithms is 100% and each of all 12 tested combinations of methods were sensitive on a single amino acid substitution. All tests have been done for 10 models of native forms of proteins of known structure (models down-loaded from the RCSB PDB 1HBB, 1HBS, 1OCC, 1U17, 1C17) and over 500 modified models; 30 known methods of prediction of secondary structure of proteins and 40 combinations of these methods included in three versions of elaborated algorithms have been examined for each protein model.