gapseq: informed prediction of bacterial metabolic pathways and reconstruction of accurate metabolic models

Table 1 Summarised comparison of CarveMe, gapseq, and ModelSEED

Metric	CarveMe	gapseq	ModelSEED
Implementation
Infrastructure	Local	Local	Web service
Input (FASTA file)	Protein	Nucleotide	Nucleotide
Programming languages	Python	Shell script, R	Perl/javascript
Gap-fill solver	CPLEX	GLPK/CPLEX	Not needed*
Gap-fill problem formulation	MILP	LP	MILP
Performance
Accuracy	0.66	0.80	0.69
Sensitivity	0.34	0.71	0.33
Specificity	0.85	0.82	0.88
Model file quality**	0.32±0.006	0.78±0.004	0.39±0.016

Accuracy, sensitivity, and specificity scores are based on 14,931 tested phenotypes including energy sources, enzyme activity, fermentation products, gene essentiality, and anaerobic food web structure predictions.
^*Solver runs on ModelSEED server. No local solver is required.
^**MEMOTE total score mean (± SD).

ISSN: 1474-760X