Decoding HMMs using the k best paths: algorithms and applications

BMC Bioinformatics

Table 3 Proteins with good labellings. Proteins with good labellings.

	τ= 0	τ= 1	τ= 2	τ= 3	τ= 4	τ= 5	Phobius
k = 1 (Viterbi)	2	11	19	27	46	59	137
k = 10	9	21	33	44	67	83	157
k = 100	34	52	64	85	110	129	198
k = 1000	46	66	89	123	143	168	214
1-best	4	18	26	38	62	79	166

For k paths and a correctness measure, we count the number of proteins for which at least one of the top k paths gives a prediction that satisfies that correctness measure. The bottom line gives the results for the 1-best decoding algorithm used in Phobius. Presented is the data set without signal peptide (n = 247).

ISSN: 1471-2105