Beyond rotamers: a generative, probabilistic model of side chains in proteins

BMC Bioinformatics

Table 1 Kullback-Leibler divergence analysis.

Amino acid	Dunbrack	Tuffery	Lovell	BASILISK backbone-dependent
Cysteine	0.01	0.03	0.00	-0.13
Aspartate	0.27	0.53	0.79	-0.49
Glutamate	0.16	0.89	1.16	-0.16
Phenylalanine	0.04	0.49	0.20	-0.35
Histidine	0.88	1.98	1.56	-0.10
Isoleucine	0.13	0.02	1.15	-0.43
Lysine	0.38	0.72	6.55	-0.23
Leucine	0.10	0.32	2.38	-0.21
Methionine	0.00	0.22	2.91	-0.06
Asparagine	0.42	0.82	3.60	-0.41
Glutamine	0.28	1.00	6.41	-0.17
Arginine	0.15	0.80	2.59	-0.17
Serine	0.04	0.16	0.10	-0.22
Threonine	0.06	0.14	0.25	-0.48
Valine	-0.07	0.13	-0.06	-0.45
Tryptophan	-0.33	0.07	1.49	-0.04
Tyrosine	-0.06	0.37	0.21	-0.33
Average	0.14	0.46	1.69	-0.28

Columns 2-4: the differences in the KL divergences between BASILISK without backbone information and a set of standard rotamer libraries. Last column: the differences in the KL divergences between backbone independent BASILISK and BASILISK including backbone information. In both cases, positive numbers indicate that backbone independent BASILISK captures the observed preferences in the test set more accurately.

ISSN: 1471-2105