Annotating functional sites in 3D biological structures

Stanford University, Stanford CA

Investigators

Linked publications & trials

Abstract

DESCRIPTION: High-throughput data collection methods have revolutionized many areas of biology and medicine. The National Library of Medicine has targeted the representation, management, and manipulation of biological structure as a key element of its mission. Following upon the success of genome sequencing and functional genomics projects, the structural biology community is creating technologies to streamline the process of determining three-dimensional biological structures--with efforts in structural genomics. Like other high-throughput efforts, a major challenge for these efforts is the appropriate annotation and indexing of structures for retrieval and analysis by biologists who are trying to understand molecular function at an atomic detail: Where are the important functional sites, and how confident are we in their location? In this proposal, we plan to develop and apply methods for annotating biological structures, so that active sites, binding sites and interaction sites in biological structures can be automatically identified and annotated. Our novel computational representation of functional sites has been successful in characterizing these sites, and recognizing them based on their biochemical and biophysical signature--a 3D motif. We propose to improve the performance of our method with basic research in the representations and algorithms used for our site models. Because our site models are manually created, our library of available models has grown slowly. We therefore further propose to accelerate the growth of our model library using a combination of supervised and unsupervised machine learning methods. First, we will use known 1D sequence motifs as "seeds" to create corresponding 3D motifs. Second, we will develop techniques for discovering entirely new motifs using cluster techniques. We will evaluate our models and resulting predictions through analysis of known structural sites, follow-up and dissemination of predictions with the structural genomics community, and large-scale evaluation on decoy and predicted structures. We will make the resulting models available on the Web for real-time structural annotation, and will distribute the software for open source development.