ANDIS: an atomic angle- and distance-dependent statistical potential for protein structure quality assessment

BMC Bioinformatics

Table 2 Performance comparison in decoy discrimination

Decoy sets	CASP5–8	CASP10–13	I-TASSER	3DRobot	Rosetta	Average^e
Dfire^b	−0.548 (2.16)	−0.441 (2.01)	− 0.480 (1.62)	−0.860 (3.77)	− 0.366 (1.97)	−0.594 (2.56)
RW	−0.550 (2.16)	−0.462 (2.01)	− 0.476 (1.59)	−0.863 (3.80)	− 0.361 (1.95)	−0.601 (2.57)
GOAP	−0.607 (2.66)	−0.550 (2.13)	− 0.473 (1.61)	−0.900 (4.04)	− 0.406 (1.99)	−0.654 (2.79)
DOOP	−0.442 (1.95)	−0.415 (1.90)	− 0.333 (1.41)	−0.874 (4.00)	− 0.285 (1.67)	−0.547 (2.51)
ITDA	−0.392 (2.03)	−0.452 (2.03)	− 0.431 (1.54)	−0.841 (3.70)	− 0.302 (1.68)	−0.545 (2.48)
VoroMQA^c	0.665 (2.66)	0.628 (2.26)	0.450 (1.44)	0.893 (3.91)	0.366 (1.86)	0.680 (2.76)
SBROD^c	0.793 (3.06)	0.831 (2.26)	0.397 (1.49)	0.857 (3.34)	0.270 (1.62)	0.741 (2.66)
AngularQA^c	0.441 (0.122)	0.426 (0.304)	0.323 (0.579)	0.543 (0.224)	0.042 (0.961)	0.422 (0.32)
ANDIS^d	−0.663 (2.80)	−0.607 (2.18)	−0.503 (1.59)	− 0.891 (3.95)	−0.401 (2.05)	− 0.681 (2.83)

^athe native structures in the decoy sets are ignored when calculating PCC and “20% enrichment”
^bThe average Pearson’s correlation coefficient between energy and TM-score (PCC) is listed outside the parentheses. The average value of 20% enrichment is listed in parentheses. “20% enrichment” means the relative occurrence of the most accurate (by TM-score) 20% models among the 20% best scoring (by potential) models compared to that for the entire decoy set. The possible value of 20% enrichment ranges from 0 to 5, the higher the better
^cSince the energy scores of VoroMQA, SBROD and AngularQA are the higher the better, the PCC between them and TM-score is positive
^dCalculation is based on a distance cutoff of 15.0 Å
^eby averaging over all 632 decoy sets

ISSN: 1471-2105