Prediction Performance in CASP4 In the recent Fourth Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction (CASP4), PROSPECT finished the 6th in the total scores in the fold recognition category out of 123 groups in the world. This is a benchmark experiment conducted every two years by groups from all over the world to test their methods and algorithms for protein structure predictions. None of the five groups that outperformed PROSPECT used threading methods, and this makes PROSPECT the best threading program in the contest. The following table summarizes the prediction performance in CASP4 based on the 32 experimental structures released so far among the 43 prediction targets. target model length segment SARF identity quality T0087 1 309 1-116 63 - good alignment (domain) T0089 1 378 200 8% good alignment T0092 1 227 148 12% good alignment T0094 1 177 18-40 28 - motif T0096 1 222 10-73 55 15% good alignment (domain) T0097 1 105 35 - unrecognized T0098 2 119 60 - good alignment T0099 1 56 43 48% perfect alignment T0100 1 342 177 - good alignment T0101 1 400 195 - fold recognized T0102 1 70 31 - fold recognized T0103 1 368 209 17% good alignment T0105 1 94 25 - unrecognized T0106 1 125 27-57 31 - motif T0107 5 188 74 - fold recognized T0108 1 179 69 - fold recognized T0109 1 182 90 - fold recognized T0110 1 95 48 - fold recognized T0111 1 429 414 50% perfect alignment T0112 1 348 294 21% perfect alignment T0114 4 87 42 - fold recognized T0115 1 296 46 - unrecognized T0116 1 759 572-713 73 - fold recognized (domain) T0120 1 203 26 - unrecognized T0121 1 372 173 27% perfect alignment T0122 1 241 216 33% perfect alignment T0123 1 160 120 61% perfect alignment T0124 1 242 999-1143 96 - fold recognized (domain) T0125 1 137 107 17% perfect alignment T0126 1 162 27 - unrecognized T0127 1 332 105 - good alignment T0128 1 211 198 50% perfect alignment Target represents CASP4 target identification number. Model gives the model number among our five submitted structures. Length is the number of amino acids in the experimental structure. Segment gives the range in the sequence when only a domain or motif is recognized by our prediction. SARF provides the number of equivalent residues for the structurally alignable portion between the experimental structure and the predicted one in a sequence independent superposition using SARF. Identity represents the sequence identity between the target and its closest homologue in PDB if it has one defined by CAFASP2 organizers (see http://www.cs.bgu.ac.il/~dfischer/CAFASP2/targets.html); otherwise a "-" is given. We have divided our prediction results into five categories: (a) perfect alignment: the predicted model and the experimental structure have a perfect alignment in the sequence-dependent superposition; (b) good alignment: the model has small errors in the alignment to the correct fold; (c) fold recognized: the correct fold is identified, but the alignment is not good; (d) motif: only a motif or a super-secondary structure is predicted correctly with a small RMSD to the corresponding portion of the experimental structure in the sequence-dependent superposition; (e) unrecognized: the model has a significantly different fold from the experimental structure.