Abstract
Abstract The amount of protein structural data is growing so rapidly that fast and accurate structure similarity search tool is in a strong demand. We have developed a structural similarity search tool SARST (Structural similarity search Aided by Ramachandran Sequential Transformation) that is able to perform extremely rapid database search with accuracy comparable to CE (Combinatorial Extension) by using a linear encoding methodology. Now we aim to modify the linear encoding strategy of SARST by integrating more protein structural information to improve its accuracy. SARST linearly encode protein structures by utilizing a Ramachandran map organized by nearest-neighbor clustering. Traditionally, Ramachandran map is a two-dimensional (2D) plot displaying the distribution of dihedral angles (φ, ψ) of residues. Different regions on this map represent different secondary structural preferences of backbone local structures; however, structural information can be lost in the process of transforming the three-dimensional (3D) protein structure into the 2D map. Our speculation is that, if we can extend the Ramachandran plot into a 3D map by adding an extra axis describing another structural property of backbone conformation, more structural information can be preserved in the transformation processes and thus improves the performance of SARST. Hence, we call the new search tool developed based on this speculation 3D-SARST. 3D-SARST, adopting the advantage of SARST, is a rapid database search tool with reasonable compromise of accuracy. Although we have not found a suitable condition to make it generally outperform SARST, we do find that 3D-SARST can achieve higher accuracy for various structural classes under specific conditions. According to the results, we can firstly determine the structural class of the query protein and then use 3D-SARST running under appropriate condition and parameter settings for that class to increase the accuracy of database searching. This two-step strategy has improved the precision of SARST by 2%, making its accuracy closer to CE. As the amount of protein structural data increases ever rapidly nowadays, we suppose that an efficient database search engine such as 3D-SARST can be valuable in many post-genomic research fields.