HMM-ModE – Improved classification using profile hidden Markov models by optimising the discrimination threshold and modifying emission probabilities with negative training sequences

Table 3 Application of HMM-d, HMM-t, HMM-ModE and HMM-Sub for function-specific classification of the S/T-Y kinase/atypical kinase/lipid kinase/ATP-grasp fold family

	HMM-d	HMM-t	HMM-ModE	HMM-Sub
2.7.1.100	19	16(0)	*	*
2.7.1.116	43	43(4)	*	*
2.7.1.117	103	103(8)	*	*
2.7.1.123	3392	529(120)	934(203)	3264
2.7.1.125	295	11(4)	11(4)	259
2.7.1.126	96	34(2)	34(2)	37
2.7.1.129	5	5(0)	*	*
2.7.1.137	135	135(29)	*	*
2.7.1.32	93	64(23)	*	*
2.7.1.38	2634	22(4)	22(4)	109
2.7.1.39	260	260(54)	*	*
2.7.1.67	57	57(15)	*	*
2.7.1.68	36	36(15)	*	*
2.7.1.82	45	31(10)	*	*
2.7.1.95	19	19(3)	*	*
2.7.9.1	171	169(55)	*	*
2.7.9.2	111	107(10)	*	*

The numbers in the parentheses for HMM-ModE and HMM-t are the total counts of sequences annotated as hypothetical, putative, unknown and unnamed which have been classified by the two protocols. A "*" in the HMM-ModE and HMM-Sub columns indicates that the number of false positive sequences picked up by the HMMER profile from the negative training data were not sufficient to build a false positive profile.

ISSN: 1471-2105