Critical Assessment of Methods of Protein Structure Prediction (CASP)-Round V John Moult, 1 Krzysztof Fidelis, 2 Adam Zemla, 2 and Tim Hubbard 3 1 Center for Advanced Research in Biotechnology, University of Maryland Biotechnology Institute, Rockville, Maryland 2 Biology and Biotechnology Research Program, Lawrence Livermore National Laboratory, Livermore, California 3 Sanger Institute, Wellcome Trust Genome Campus, Cambridgeshire, United Kingdom ABSTRACT This article provides an introduc- tion to the special issue of the journal Proteins dedicated to the fifth CASP experiment to assess the state of the art in protein structure prediction. The article describes the conduct, the categories of pre- diction, and the evaluation and assessment proce- dures of the experiment. A brief summary of progress over the five CASP experiments is provided. Related developments in the field are also described. Proteins 2003;53:334 –339. © 2003 Wiley-Liss, Inc. Key words: protein structure prediction; communi- tywide experiment; CASP INTRODUCTION This issue of Proteins is devoted to articles reporting the outcome of the fifth communitywide experiment to assess methods of protein structure prediction (CASP5) and related activities. Four previous CASP experiments in 1994, 1996, 1998, and 2000 were reported in previous special issues of Proteins 1–4 as well as elsewhere. 5–13 Independent discussions of CASP5 have also appeared. 14,15 The primary goals of CASP are to establish the capabili- ties and limitations of current methods of modeling protein structure from sequence, to determine where progress is being made, and to determine where the field is held back by specific bottlenecks. With 10 years of effort now recorded, these latter factors—progress or the lack of it— have assumed increasing importance. Methods are assessed on the basis of the analysis of a large number of blind predictions of protein structure. This article outlines the structure and conduct of the experiment and is followed by a description of the CASP5 target proteins. There are sections of the special issue for each of the main CASP prediction categories: Comparative Modeling, Fold Recognition, and New Fold methods. These sections begin with an article by the assessment team in that area and continue with contributions from the predic- tion groups the assessors considered to have done the most interesting work. The number of predictor articles in each category varies—there are only three in comparative modeling, four in the fold recognition, and seven in the new folds category. The small number of articles in compara- tive modeling reflects the fact that there again appears to have been little progress in this area since the last CASP. The assessors’ articles are probably the most important in the whole issue and describe the state of the art as they found it in CASP5. The role and importance of automated servers in the structure prediction field continue to grow. Another main section of the issue deals with this topic. The first of these articles describes the CAFASP3 experiment. The goal of CAFASP is to assess the state of the art in automatic methods of structure prediction. 16 Whereas CASP allows any combination of computational and human methods, CAFASP captures predictions directly from fully auto- matic servers. CAFASP makes use of the CASP target distribution and prediction collection infrastructure, but is otherwise independent. The results of the CAFASP3 experi- ment were also evaluated by the CASP assessors, provid- ing a comparison of fully automatic and hybrid methods. Full information is available at the Web site (http:// www.cs.bgu.ac.il/dfischer/CAFASP3/). This first article is followed by three that report some of the more interest- ing CAFASP results. Large-scale benchmarking of prediction server perfor- mance is reported in the following two articles: one for Livebench, 16 and one for EVA. 17 In contrast to CASP and CAFASP, the benchmarking experiments run continu- ously. Both Livebench and EVA operate by sending the sequences of just released PDB entries to automatic predic- tion servers and collating and analyzing the results over time. Livebench focuses primarily on fold recognition and EVA primarily on secondary structure predictions. Live- bench and EVA are entirely independent of CASP, and we are grateful to the organizers for their participation in the CASP5 meeting and in contributing to this issue of Pro- teins. Benchmarking experiments complement CASP, par- ticularly by clarifying issues of the statistical significance of the results. Prediction of disorder in protein structures was included for the first time in CASP5. An assumption behind the prediction of protein structure is that, under specified environmental conditions, every protein molecule has es- Grant sponsor: Lawrence Livermore Laboratory; Grant sponsor: Department of Energy; Grant sponsor: National Library of Medicine; Grant number: LM07085; Grant sponsor: National Institutes of Health; Grant number: GM/DK61967; Grant sponsor: HP; Grant sponsor: Amgen; Grant sponsor: GlaxoSmithKline; Grant sponsor: IBM. *Correspondence to: John Moult, Center for Advanced Research in Biotechnology, University of Maryland Biotechnology Institute, 9600 Gudelsky Dr., Rockville, MD 208.50, E-mail: moult@umbi.umd.edu Received: 21 June 2003; Accepted 24 June 2003 PROTEINS: Structure, Function, and Genetics 53:334 –339 (2003) © 2003 WILEY-LISS, INC.