Short-read reading-frame predictors are not created equal: sequence error causes loss of signal

BMC Bioinformatics

Table 1 Running times per gigabase of sequence data on a single 2 GHz processor

Tool	Method	Symbol	Ref.	Time/Gbase
FragGeneScan	Hidden Markov Model	FGS3,FGS5	[11]	6 hours
MetaGeneAnnotator	Codon usage + start site heuristics	MGA	[9]	15 min
MetaGeneMark	Codon usage + gc-content heuristics	MGM	[8]	20 min
Orphelia	Neural network	OPH	[10]	13 hours
Prodigal	Codon usage + dynamic programming	PRD	[12]	30 min

Compared with downstream analyses, ab initio gene calling is computationally inexpensive.

ISSN: 1471-2105