Table 1.

Docking benchmark using different 3D models of mOR256-3 and 52 compounds

Model	MD snapshota	MCC	Hit rateb(precision)	Recallc	True positive	True negative	False positive	False negative
In-house model	1	0.50	0.60	0.60	6	38	4	4
	2	0.26	0.40	0.40	4	36	6	6
In-house model without ECL2	1	0.26	0.40	0.40	4	36	6	6
	2	0.13	0.30	0.30	3	35	7	7
SWISS-MODEL	1	0.23	0.36	0.40	4	35	7	6
	2	0.20	0.33	0.40	4	34	8	6
AlphaFold 2	1	0.38	0.50	0.50	5	37	5	5
	2	0.26	0.40	0.40	4	36	6	6

^aTwo snapshots that gave the best Matthew’s correlation coefficient (MCC) as a statistical measure of the model’s predictivity (64). MCC returns a value between −1 (total disagreement between prediction and observation) and +1 (perfect prediction).

^bHit rate or precision, the fraction of true ligands among the model predicted ones.

^cRecall indicates the fraction of true ligands retrieved by the model out of all the true ligands in the benchmark compounds.