Figure 1

The FRAGlib fragments database is build from ASTRAL representative subset of SCOP database using 40% sequence similarity threshold (see right picture). We store the symbolized local structure representation codes of each fragment together with the homology sequence profile (see left picture). Both are dissected from the SLSR codes and homology profile of a parent protein. The string of SLSR codes representing the local structure of the Cα chain in the phi-phi space. We remove from the fragments database all identical in terms of both SLSR codes and sequence homology profile fragments. On the left picture we present the FRAGlib module for prediction of local structural segments using homology profile similarity and the fragments database. The Query protein is dissected into short parts (from 7 up to 19 residues long). For each part the similarity search is performed. Any member of the fragments database which is similar in terms of homology sequence profile similarity is added to the list of predicted structures for this short part of query protein. This list is then sorted and cut after arbitrary chosen 20^th position. If the highest score of predicted fragments is below the user's cut-off value whole prediction is discarder. In the end some of parts of a query protein are covered by list of 20 fragments from the database. They are called the predicted local structural segments (PLSSs).