Protein Binding Ligand Prediction Using Moments-Based Methods

Structural genomics initiatives have started to accumulate protein structures of unknown function in an increasing pace. Conventional sequence-based function prediction methods are not able to provide useful function information to most of such structures. Thus, structure-based approaches have been developed, which predict function of proteins by capturing structural characteristics of functional sites. Particularly, several approaches have been proposed to identify potential ligand binding sites in a query protein structure and to compare them with known ligand binding sites. In this chapter, we introduce computational methods for describing and comparing ligand binding sites using two dimensional and three dimensional moments. An advantage of moment-based methods is that the tertiary structure of pocket shapes is described compactly as a vector of coefficients of series expansion. Thus a search against an entire PDB-scale database can be performed in real-time. We evaluate two binding pocket representations, one based on two-dimensional pseudo-Zernike moments and the other based on threedimensional Zernike moments. A new development of pocket comparison method is also mentioned, which allows partial matching of pockets by using local patch descriptors.