Combined Density- and Grid- Based Method for Clustering of Protein Substructures

Cluster analysis, data mining, PDB, sequence-structure fragments, protein structure prediction

Data mining techniques may reveal interesting knowledge in various datasets. The biological databases are enormously large and therefore, data mining techniques could be extremely helpful to extract the knowledge from them. In our study, we focused on data mining in PDB - Protein Data Bank. We used cluster analysis to identify the sequences that occur in limited number of structural conformations (sequence-structure fragments). This knowledge about protein fragments can be used in protein structure predictions. In this paper, we present a combined density- and grid-based method that we developed for clustering of protein structures. We also compare this method with a simple density-based clustering method that we used in the first part of our study to prove the existence of protein sequences that occur in more than one structural conformation but the number of its structural conformations is limited.