How Good Are Simplified Models for Protein Structure Prediction?

Table 2

Proportion of the number of protein sequences that fall in the different ranges of correlation coefficient between energy function value and distant root mean square deviation (dRSMD), produced by different sampling techniques: sampled by PCLF search, sampled by energy function guidance, and sampled by random walk.

Energy function	Correlation coefficient for cubic lattice (%)				Correlation coefficient for FCC lattice (%)
	>0	>0.5	≤0	<−0.5	>0	>0.5	≤0	<−0.5
Sampling by using PCLF search
hp-basic	11.56	2.02	88.43	58.67	5.73	0.28	94.26	74.21
mj	70.20	24.35	29.79	6.59	73.72	28.85	26.28	6
bre	9.16	0.57	90.83	66.47	3.71	0	96.28	80
Sampling by using guided search
hp-basic	9.88	5.14	90.12	79.05	3.69	0.86	96.31	90
mj	31.85	16.38	68.15	51.63	19.31	5.42	80.69	51.84
bre	2.09	1.43	97.91	93.98	0.86	0.22	99.14	93.49
Sampling by using random walk
hp-basic	86.22	1.04	13.78	0.13	87.54	1.23	2.36	0.11
mj	31.76	0.13	68.24	0	54.25	1.18	46.75	0.24
bre	85.30	2.62	14.70	0.26	81.28	1.28	18.72	1.54