Combining metabolome and clinical indicators with machine learning provides some promising diagnostic markers to precisely detect smear-positive/negative pulmonary tuberculosis

Table 3 Classification performance of binary classifications with selected feature combinations on test sets

	Accuracy	Sensitivity	Specificity	PPV	NPV
RF
SPPT/Ctrl (F₁)	83.33%	80.00%	85.71%	80.00%	85.71%
SNPT/Ctrl (F₂)	92.86%	85.71%	100%	100%	87.50%
SPPT/SNPT (F₃)	83.33%	80.00%	85.71%	80.00%	85.71%
SVM
SPPT/Ctrl (F₁)	91.67%	80.00%	100%	100%	87.50%
SNPT/Ctrl (F₂)	92.86%	85.71%	100%	100%	87.50%
SPPT/SNPT (F₃)	91.67%	80.00%	100%	100%	87.50%
MLP
SPPT/Ctrl (F₁)	83.33%	60.00%	100%	100%	77.78%
SNPT/Ctrl (F₂)	92.86%	85.71%	100%	100%	87.50%
SPPT/SNPT (F₃)	91.67%	80.00%	100%	100%	87.50%

F₁ set: albumin and 9-OxoODE; F₂ set: l-Pyroglutamic acid, Enterostatin human and 9-OxoODE; F₃ set: Val-Ser, Methoxyacetic acid and Ethyl 3-hydroxybutyrate

ISSN: 1471-2334