Table 2.

Description of candidate blood transcriptional signatures for tuberculosis

	Original gene number	Model	Intended application	Discovery datasets
				Population	HIV status	Setting	Approach	Tuberculosis cases	Controls	Total
Anderson39.LTBI26	42	Disease risk score	Tuberculosis vs LTBI	Children	Positive or negative	South Africa, Malawi	Elastic net using genome-wide data	87	43	130
Anderson39.OD26	51	Disease risk score	Tuberculosis vs OD	Children	Positive or negative	South Africa, Malawi	Elastic net using genome-wide data	87	134	221
BATF227	1	NA	Tuberculosis vs HC (acute vs convalescent)	Adults	Negative	UK	SVM using genome-wide data	46	31	77
Duffy1016	10	SVM (linear kernel)	Tuberculosis vs LTBI and OD	Adults	Positive or negative	South Africa	Multinomial random forest using genome-wide data	93	207	300
Gjoen828	7	LASSO regression	Tuberculosis vs HC and OD	Children	Negative	India	LASSO using 198 pre-selected genes	47	36	83
Gliddon329	3	(FCGR1A + C1QB) − (ZNF296)	Tuberculosis vs LTBI	Adults	Positive or negative	South Africa, Malawi	FS-PLS using genome-wide data	NS	NS	285
Gliddon429	4	(GBP6 + PRDM1) − (TMCC1 + ARG1)	Tuberculosis vs OD	Adults	Positive or negative	South Africa, Malawi	FS-PLS using genome-wide data	NS	NS	293
Huang1130	13	SVM (linear kernel)	Tuberculosis vs HC and OD	Adults	Negative	UK	Common genes from elastic net, L1/2 and LASSO models, using genome-wide data	16	79	95
Kaforou2531	27	Disease risk score	Tuberculosis vs LTBI	Adults	Positive or negative	South Africa, Malawi	Elastic net using genome-wide data	NS	NS	285
Kaforou3931	44	Disease risk score	Tuberculosis vs OD	Adults	Positive or negative	South Africa, Malawi	Elastic net using genome-wide data	NS	NS	293
Kaforou4531	53	Disease risk score	Tuberculosis vs LTBI and OD	Adults	Positive or negative	South Africa, Malawi	Elastic net using genome-wide data	NS	NS	NS
Maertzdorf432	4	Random forest	Tuberculosis vs HC	Adults	Negative	India	Random forest using 360 selected target genes	113	76	189
NPC233	1	NA	Tuberculosis vs HC and LTBI	Adults	NS	Brazil	Differential expression using genome-wide data	6	28	34
Penn-Nicholson617	6	Difference of means	Incipient tuberculosis vs HC	Adolescents	Negative	South Africa	SVM-based gene pair models using genome-wide data	46	107	153
Qian1734	17	Sum of standardised expression	Tuberculosis vs HC and OD	Adults	Negative	UK	Differential expression of Nrf2-mediated genes	16	69	85
Rajan535	5	Unsigned sums	Tuberculosis vs HC (screening among PLHIV)	Adults	Positive	Uganda	Differential expression using genome-wide data	NS	NS	80 (1:2 cases:controls)
Roe313	3	SVM (linear kernel)	Incipient tuberculosis vs HC	Adults	Negative	UK	Stability selection using genome-wide data	46	31	77
Roe427	4	SVM (linear kernel)	Tuberculosis vs OD	Adults	Negative	UK	SVM using genome-wide data	23	35	58
Roe527	5	SVM (linear kernel)	Tuberculosis vs HC and OD	Adults	Negative	UK	SVM using genome-wide data	23	50	73
Singhania2036	20	Modified disease risk score	Tuberculosis vs HC and OD	Adults	Negative	UK, South Africa	Random forest using modular approach	NS	NS	NS
Suliman237	2	ANKRD22 −OSBPL10	Incipient tuberculosis vs HC	Adults	Negative	The Gambia, South Africa	Pair ratios algorithm using genome-wide data	79	328	407
Suliman437	4	(GAS6 + SEPT4) – (CD1C + BLK)	Incipient tuberculosis vs HC	Adults	Negative	The Gambia, South Africa, Ethiopia	Pair ratios algorithm using genome-wide data	45	141	186
Sweeney338	3	(GBP5 + DUSP3)/2 −KLF2	Tuberculosis vs LTBI and OD	Adults	Positive or negative	Meta-analysis of South Africa, Malawi, UK, France, USA	Significance thresholding and forward search in genome-wide data	296	727	1023
Walter4639	51	SVM (linear kernel)	Tuberculosis vs LTBI	Adults	Negative	USA	SVM using genome-wide data	24	24	48
Walter3239	47	SVM (linear kernel)	Tuberculosis vs OD	Adults	Negative	USA	SVM using genome-wide data	24	24	48
Walter10139	119	SVM (linear kernel)	Tuberculosis vs LTBI and OD	Adults	Negative	USA	SVM using genome-wide data	24	48	72
Zak1640	16	SVM (linear kernel)	Incipient tuberculosis vs HC	Adolescents	Negative	South Africa	SVM-based gene pair models using genome-wide data	37	77	114

Signatures were identified by systematic literature review and included for analysis. Signature names represent the first author's name of the corresponding publication, suffixed with the number of constituent genes that are present in the current RNAseq dataset. Both Anderson signatures resulted in the same number of final genes; these signatures were therefore additionally appended with the comparator control group. Details on how models were recreated are in appendix 1 (pp 2-4). LTBI=latent tuberculosis infection. OD=other diseases. NA=not applicable. HC=healthy controls. SVM=support vector machine. LASSO=least absolute shrinkage and selection operator. FS-PLS=forward selection-partial least squares. NS=not specified. Nrf2=nuclear factor, erythroid 2-like 2. PLHIV=people living with HIV.