Table 1. Fold recognition over the test decoy sets.

	number of		basic method		DC method		DC best
PDB	decoy models		true folds		true folds		top model
code	basic	DC method	top 5	top 10	top 5	top 10	RMS (Å/ca)
2trx	16768	12526 (5046)	0	0	5 (1)	8 (4)	6.55/105
3chy	7015	700 (388)	1	2	5 (5)	10 (8)	5.70/121
1f4p	4243	3896 (991)	0	0	3 (5)	6 (8)	6.73/134
5p21	20169	5196 (1655)	2	4	4 (4)	4 (4)	8.70/158

Of the five proteins considered, one (1cozA) did not produce models and is not tabulated. For each of the other proteins (PDB) the number of decoys constructed by the basic PLATO method is tabulated in the leftmost column along with those using the PLATO method with direct contact (DC) information and, in parenthes, the number after applying structural constraints to the contact matrix. The computer execution time is roughly proportional to these numbers. The number of true folds (defined by topology string) found in the top 5 and top 10 ranked positions is tabulated for the basic PLATO method and the DC augmented method ranked by the PLATO score and the DC score combined as their geometric mean as used previously (with the arithmetic mean in parenthes). The number of hits over larger subsets is more easily seen in the plots in Figure 5. In the rightmost column, the root mean square (RMS) deviation was calculated over the number of residues (CA atoms) shown in parentheses for the top model. These values are slightly higher than some reported previously as the current models were not selected using residue-level contact data.