Intelligent decision support for protein crystal growth by I. Jurisica P. Rogers J. I. Glasgow S. Fortier J. R. Luft J. R. Wolfley M. A. Bianca D. R. Weeks G. T. DeTitta Current structural genomics projects are likely to produce hundreds of proteins a year for structural analysis. The primary goal of our research is to speed up the process of crystal growth for proteins in order to enable the determination of protein structure using single crystal X-ray diffraction. We describe Max, a working prototype that includes a high- throughput crystallization and evaluation setup in the wet laboratory and an intelligent software system in the computer laboratory. A robotic setup for crystal growth is able to prepare and evaluate over 40 thousand crystallization experiments a day. Images of the crystallization outcomes captured with a digital camera are processed by an image-analysis component that uses the two-dimensional Fourier transform to perform automated classification of the experiment outcome. An information repository component, which stores the data obtained from crystallization experiments, was designed with an emphasis on correctness, completeness, and reproducibility. A case-based reasoning component provides support for the design of crystal growth experiments by retrieving previous similar cases, and then adapting these in order to create a solution for the problem at hand. While work on Max is still in progress, we report here on the implementation status of its components, discuss how our work relates to other research, and describe our plans for the future. P roteins are involved in every biochemical pro- cess that maintains life in a living organism. One of the fundamental challenges of modern molecu- lar biology is discovering the laws that control how proteins evolve their three-dimensional structure. Through an increased understanding of protein structure we can gain insight into the functions of these important molecules. Currently, the most pow- erful method for determining protein structure is sin- gle crystal X-ray diffraction. A crystallography experiment begins with a crystal that ideally diffracts X-rays to high resolution, i.e., it produces a high-quality diffraction pattern that re- veals the crystal’s internal order. Crystals are reg- ular, repeating arrays of atoms or molecules in three- dimensional space. The basic building block of a crys- tal is called a unit cell, the smallest unit of a lattice de- fined by three axes and the three angles between them. In order for a protein crystal to diffract at high reso- lution, it should not have large unit cell dimensions. Determining protein structure is often limited by the difficulty of growing crystals suitable for diffraction. This is partially due to the large number of param- eters affecting the crystallization outcome (e.g., pu- rity of proteins, intrinsic physico-chemical, biochem- ical, biophysical, and biological parameters) and the unknown dependencies between the variation of these parameters and the propensity of a given mac- romolecule to crystallize. The primary goal of the research described in this paper is to develop a com- prehensive repository of data from crystal growth ex- periments (both successful and unsuccessful) and ap- ply this knowledge in an intelligent decision-support system for planning novel experiments. Copyright 2001 by International Business Machines Corpora- tion. Copying in printed form for private use is permitted with- out payment of royalty provided that (1) each reproduction is done without alteration and (2) the Journal reference and IBM copy- right notice are included on the first page. The title and abstract, but no other portions, of this paper may be copied or distributed royalty free without further permission by computer-based and other information-service systems. Permission to republish any other portion of this paper must be obtained from the Editor. JURISICA ET AL. 0018-8670/01/$5.00 © 2001 IBM IBM SYSTEMS JOURNAL, VOL 40, NO 2, 2001 394