Complementing sequence-derived features with structural information extracted from fragment libraries for protein structure prediction

Table 1 Overall performance of protein folding with potentials derived from fragment libraries

Test set	Potentials	Avg. TM of Best (± s.d.)	Avg. TM of Top1 (± s.d.)	# Best with TM > 0.5	# Top1 with TM > 0.5
CASP13 FM	trRosetta Cβ dist	0.5523 (± 0.1154)	0.4572 (± 0.1547)	22	15
CASP13 FM	trRosetta Cβ dist + FragLib wGMM	0.5689 (± 0.1082)	0.4660 (± 0.1647)	23	15
CASP13 TBM	trRosetta Cβ dist	0.6619 (± 0.1069)	0.4806 (± 0.2119)	53	26
CASP13 TBM	trRosetta Cβ dist + FragLib wGMM	0.6610 (± 0.1098)	0.5806 (± 0.1800)	52	43
CAMEO	trRosetta Cβ dist	0.5932 (± 0.1326)	0.4953 (± 0.1779)	101	70
CAMEO	trRosetta Cβ dist + FragLib wGMM	0.6011 (± 0.1322)	0.5212 (± 0.1765)	102	78
Overall	trRosetta Cb dist	0.6051 (± 0.1279)	0.4862 (± 0.1839)	176	111
Overall	trRosetta Cβ dist + FragLib wGMM	0.6119 (± 0.1278)	0.5292 (± 0.1788)	177	135

Potentials derived from fragment libraries were utilized to predict protein structures with trRosetta distance potential as a baseline. For each target, the top1 decoy with lowest energy and the best decoy with the highest TM-Score were picked up for evaluation. The mean and standard deviation of TM-Scores of all targets and the number of selected decoys with TM-Score > 0.5 were evaluated. The better performance in each category is highlighted in bold

ISSN: 1471-2105