TABLE 8.

Validating the effect of including premature stop codons and excluding non-predictive mutations from the mutation library.

Drug	Res	Sus	Initial mutations			Including prem. stop codons			Excluding non-predictive muts
			Spec.	Sens.	MCC	Spec.	Sens.	MCC	Spec.	Sens.	MCC
RMP	596	1814	0.978	0.896	0.885	0.978	0.896	0.885	0.986	0.886	0.895
INH	768	1641	0.946	0.879	0.826	0.945	0.880	0.826	0.969	0.870	0.854
STM	379	687	0.85	0.744	0.592	0.817	0.805	0.606	0.902	0.778	0.688
EMB	304	880	0.739	0.914	0.576	0.739	0.914	0.576	0.825	0.908	0.666
PZA	239	420	0.971	0.799	0.802	0.971	0.808	0.809	0.971	0.808	0.809
KAN	211	176	0.920	0.896	0.814	0.920	0.896	0.814	0.920	0.896	0.814
FLQ	271	296	0.905	0.878	0.784	0.905	0.878	0.784	0.909	0.878	0.788
AMK	213	278	0.953	0.850	0.814	0.953	0.850	0.814	0.964	0.826	0.807
ETH	184	245	0.624	0.853	0.479	0.539	0.919	0.479	0.539	0.919	0.479
CAP	191	261	0.877	0.864	0.738	0.877	0.864	0.738	0.877	0.864	0.738

PointFinder-predicted resistance was compared to drug susceptibility test results on the validation data set. The quality of the prediction was evaluated with three different mutation libraries; the initial mutation library (initial mutations). The mutation library including premature stop codons in katG, pncA, ethA, and gidB (including prem. stop codons), and the mutation library both including those premature stop codons and excluding mutations considered non-predictive for resistance (excluding non-predictive muts). Green indicates an increase and red a decrease in the performance measurement compared to same measurement from the mutation library to the left. Res, number of resistant strains determined by drug susceptibility testing; Sus, number of susceptible strains determined by drug susceptibility testing; Spec, specificity; Sens, sensitivity; MCC, Mathew Correlation Coefficient; RMP, Rifampicin; INH, Isoniazid; STM, Streptomycin; EMB, Ethambutol; PZA, Pyrazinamide; KAN, Kanamycin; FLQ Fluoroquinolone; AMK, Amikacin; ETH, Ethionamide; CAP, Capreomycin.