A New Biclustering Algorithm with Exclusive Random Selection of Columns for Predicting Recognition Spots on Protein Molecular Surfaces

Abstract

Abstract:A protein and a ligand have a process of recognizing each other when they are remote before approaching for binding. In this process, there should be a local portion corresponding to a target of the recognition (a recognition spot) on the protein molecular surface. In this paper, we proposed a new biclustering algorithm BISERS for predicting recognition spots on protein molecular surfaces, in which a portion of the molecular surface of a query protein that frequently shows the similarity to other specific proteins binding to the similar ligand is extracted by biclustering. In BISERS, the similarity between rows as well as the similarity between columns is introduced into the evaluation function for updating the biclusters. In addition, the random sampling is applied to the process of exclusive selection of the column for reducing the computational cost effectively. Experimental results for 60 proteins show the effectiveness of our BISERS algorithm.

Extraction of Feature Values using Protein Molecular Surface Data and Prediction of Significant Spot for Preferential Binding of Ligands

“…6. Ligand binding site "HETATM" data (12) . method, the number of feature points extracted by LBE2 is small.…”

“…As a result, with respect to feature points representing two pockets, it is possible to obtain a set of feature point pairs in high speed that can be regarded as similar among the pockets. With a general-purpose CPU, we compare sets of feature points based on feature value and generate a list of similar feature points corresponding to 3D space (12) .…”

“…Therefore, for a set of feature points that constitute a protein pocket, a set of feature points similar to them is extracted from many pockets, and biclustering processing is performed to realize extraction of the significant spot. As a method of biclustering to predict the significant spot, we use BISERS (12) proposed by Nishimura et al…”

“…A total of 60 × 59 matching processes compare 60 types of proteins with other proteins and perform biclustering processing using the results. In order to show the effectiveness of the proposed method, we compared it with the method of Nishimura et al (12) . In the method of Nishimura et al, feature points are extracted by a curvature-based approach (10) , and comparison is performed between feature points using point feature histograms (13) as feature quantities.…”