Applications of AI to Protein Crystallography

Dr. Thomas Ioerger Department of Computer Science Texas A&M University

4:10pm Wed, Sept. 15, 1999 Rm 124, HRBB

Abstract

Proteins are important macromolecules that serve a wide variety of biological functions in cells. Knowledge of the structures of proteins is useful for elucidating their mechanisms, determining their evolutionary relationships, understanding the molecular basis of disease, and designing drugs (e.g. inhibitors) that interact with them. One of the most important methods for determining the structures of proteins in a laboratory is X-ray crystallography. Crystallography has many complex steps which culminate in the production of a 3D map of "cloud-like" electron density representing the protein. However, the final phase of interpreting this electron density map and building a 3D model of the protein structure (with coordinates of atoms) still remains one of the most challenging steps to automate, and has become a major impediment to progress in structural biology.

We have developed a system called TEXTAL that uses pattern recognition and other AI methods to automatically interpret electron density maps and thereby solve protein structures. Given a spherical region of density in a map to be solved, TEXTAL uses a nearest-neighbor algorithm to look up regions with similar patterns of electron density in a database of previously solved maps. This procedure crucially relies on the extraction of rotation-invariant features that help characterize and match these patterns of density. Then atomic coordinates from the known structures in matched regions are rotated and translated into position to incrementally build the model for the new map. TEXTAL has already been demonstrated to build fairly accurate models from real electron density maps without human intervention, and many further improvements are currently being investigated. Our experiments with TEXTAL validate the utility of pattern recognition for interpreting electron density maps to solve protein structures, and have revealed a number of related opportunities for using AI methods to incorporate human expertise into a system for automating this complex task. This project is a collaboration with Dr. James C. Sacchettini in the Department of Biochemistry at TAMU, and is funded by the NIH.