'Structure-based drug design' involves a rational approach for drug discovery by understanding the three dimensional structural interactions between the target protein and the ligands. The central assumption of structure-based design is that good inhibitors must possess significant structural and chemical complementarity to their target receptor. This necessitates a scientific analysis of the currently available protein-ligand complexes to identify the critical interactions in the binding of the ligands to the receptor.

With recent advances in high throughput experimental techniques and easy availability of high computational power, the amount of protein-ligand structural data generated has explosively increased in the last decade. These structural data could be either from experimental methods such as X-ray crystallography and NMR, or from in silico approaches such as structure-based docking. However, the ability to fully leverage this information in drug discovery depends on the ability to visualize, organize, analyze, and mine the data. Structure Interaction Fingerprint (SIFt) is a tool developed by Deng et al. [1] at Biogen-Idec, for representing and analyzing 3-D protein-ligand interactions. SIFt involves representation of the 3D structural binding information between a protein and a ligand as a one-dimensional bit-string, after initial identification of all the residues and the interactions involved in the binding. By defining a similarity measure between SIFts, one can cluster them, perform binding-mode analysis, or examine interactions across multiple members of family. The current version of SIFt however is still elementary and its applicability limited, and there is scope for further development of SIFt to extend its usability.

A weighted version of SIFt, called w-SIFt, will be presented in the talk. The weighting extends the capability of SIFt to rank order or classify inhibitors of a target protein in accordance to their potency (or any other biochemical assay value, to be general). It is shown that w-SIFt involves assigning weights to residues based on their relative importance in the binding. We applied Non-negative Matrix Factorization (NMF), a dimensionality reduction method that involves parts based learning of objects, to identify the inherent linearly independent features embedded in the SIFt data. The weights of the SIFt bits are determined by optimizing a nonlinear objective function using Simulated Annealing. The mathematical details involved in determining the weights of w-SIFt will be discussed during the presentation.

1. Deng Z, Chuaqui C, Singh J. Structural interaction fingerprint (SIFt): a novel method for analyzing three-dimensional protein-ligand binding interactions. J Med Chem. 2004 Jan 15; 47(2):337-44.