Genomic prediction of tuberculosis drug-resistance: benchmarking existing databases and prediction algorithms

BMC Bioinformatics

Table 2 Summary of isolates in the three TB databases according to the 14 anti-TB drugs

Drug	PATRIC				ReSeqTB				LitRev
Drug	N _clean	R _error	N _full	R _missing	N _clean	R _error	N _full	R _missing	N _clean	R _error	N _full	R _missing
INH	5018	0.043	5018	0.000	3503	0.045	3555	0.015	5451	0.043	5451	0.000
RIF	4979	0.028	4981	0.000	3459	0.038	3521	0.018	5416	0.034	5416	0.000
EMB	4739	0.075	4787	0.010	3497	0.098	3546	0.014	5335	0.092	5335	0.000
PZA	3633	0.054	3633	0.000	3298	0.095	3346	0.014	4752	0.082	4752	0.000
STM	3367	0.141	3380	0.004	1951	0.129	2008	0.028	3716	0.129	3716	0.000
AMK	1131	0.034	1131	0.000	983	0.040	993	0.010	1256	0.037	1256	0.000
CAP	1100	0.055	1101	0.001	1158	0.076	1167	0.008	1587	0.059	1588	0.001
KAN	1348	0.056	1350	0.001	716	0.031	716	0.000	1095	0.031	1095	0.000
CIP	340	0.018	340	0.000	358	0.031	358	0.000	313	0.016	313	0.000
MFX	726	0.059	726	0.000	874	0.071	885	0.012	993	0.057	993	0.000
OFX	836	0.065	851	0.018	1163	0.070	1188	0.021	1818	0.052	1818	0.000
ETO	559	0.370	562	0.005	252	0.159	252	0.000	321	0.265	321	0.000
PTO	52	0.115	52	0.000	410	0.337	431	0.049	498	0.301	498	0.000
PAS	375	0.208	375	0.000	78	0.026	78	0.000	74	0.108	74	0.000

N_clean refers to the number of isolates with valid DST results and genetically-inferred credibility scoring; R_error refers to the misclassification rate in each database, defined as the proportion of the N_clean isolates with DST credibility scores < 0.5; N_full is defined as the summation of N_clean and the number of good mapping quality isolates with no genetically-inferred credibility scoring; and R_missing refers to the proportion of N_full isolates that presented unusable laboratory-based DST phenotypes due to either inconsistent results (across multiple DST phenotype entries for the same isolate) or documentation errors across the databases

ISSN: 1471-2105