A modified decision tree approach to improve the prediction and mutation discovery for drug resistance in Mycobacterium tuberculosis

Table 1 Predictive performance across algorithms

Drug	Total tests	% resistance	TB-Profiler				Treesist-TB^a
Drug	Total tests	% resistance	Sens	Spec	Acc	AUC	Sens	Spec	Acc	AUC
INH	1835	16.2	86.2	98.4	96.5	92.3	84.2	99.2	96.8	91.7
RIF	2045	8.1	90.3	98.2	97.6	94.2	86.1	98.5	97.5	92.3
EMB	1999	3.5	71.4	96.7	95.8	84.1	57.1	98.2	96.8	77.7
PAS	1114	8.8	38.8	95.7	90.7	67.2	64.3	90.6	88.2	77.4
CYS	833	18.0	30.7	95.2	83.6	62.9	45.3	93.7	85.0	69.5
ETH	2118	32.2	71.1	78.6	76.2	74.8	72.1	75.8	74.6	73.9
			*Regular Decision Tree*				Single optimized Tree^b
			Sens	Spec	Acc	AUC	Sens	Spec	Acc	AUC
INH	1835	16.2	85.6	100	97.7	92.9	80.2	99.2	96.1	89.8
RIF	2045	8.1	81.2	100	98.5	91.5	87.3	99.8	98.8	93.6
EMB	1999	3.5	32.9	99.7	97.3	82.9	34.3	99.5	97.2	83
PAS	1114	8.8	64.3	100	96.9	85.5	50	97.8	93.6	74.1
CYS	833	18.0	33.3	99.4	87.5	67.3	35.3	98	86.7	66.7
ETH	2118	32.2	48.8	94.3	79.7	77.5	49.6	92.5	78.7	76.2

INH Isoniazid, RIF Rifampicin, PAS para-aminosalisylic acid, CYS cycloserine, ETH ethionamide, EMB Ethambutol, Sens Sensitivity, Spec Specificity, Acc Accuracy, AUC Area under the ROC Curve
^adefault application of Treesist-TB
^bapplication of Treesist-TB with a single combined study dataset

ISSN: 1471-2164