Our results at CAPRI

CAPRI (Critical Assessment of Predicted Interactions), the first communitywide experiment devoted to protein docking (Janin et al., 2003). For each target, predictor groups may submit up to ten models to independent assessors, who compare the predictions with newly obtained X-ray structures of the complexes, which crystallographers make available for the evaluation. So far, CAPRI results were evaluated for 28 targets in Rounds 1-12. Two targets (22 and 23) were cancelled due to the early release of X-ray structures, and no acceptable solutions were submitted for three targets (4, 5, and 28), leaving 23 valid targets. The table here shows the summary of results for all groups that submitted at acceptable or better solutions for 10 or more targets in Rounds 1-12 of CAPRI. The table was obtained by summing the results for each group from the three separate CAPRI evaluation meetings. We have participated both as a predictor group (in Rounds 1-5 under the name of Camacho, who was a research assistant professor in the PI’s lab at that time) and as the ClusPro server (starting with Round 3). As shown, we have submitted acceptable or better predictions for 12 of the 23 valid targets, which amounts to the second best overall performance in CAPRI Rounds 1-12. The table also shows the performance of the ClusPro server, which had to submit predictions within 24 hours and without the use of any biological information.

Best performing predictor groups in Rounds 1-12
Predictor group	*	**	***
Weng (Boston University)	14	7	3
Camacho/Vajda (Boston U.)	12	5	4
Abagyan (Scripps)	11	6	3
Baker (Univ. of Washington)	11	4	5
Wolfson/Nussinov (Tel Aviv U.)	11	3	1
Eisenstein (Weizmann Inst.)	10	1	4
ClusPro (Boston University)	7	2	1
*Acceptable, ** Medium, *** High accuracy prediction as defined by Mendez et al (2003)