Table 4.

Mean results of ten repeats of 5-fold cross-validation on (b) the full data set and (c) suspected diagnosis data set, compared to traditional screening results (a). Methods are applied to all 53 features, 5 features selected with ANOVA or LDA dimensions. The methods were evaluated by sensitivity, specificity, and number of false positives (real numbers are rounded up) combined for training and test set. For the suspected diagnosis data set, these evaluations are calculated based on the full data set to allow comparability between both data sets. Abbreviations: Asa—argininosuccinate, C14:1—tetradecenoylcarnitine, C14OH—3-OH-tetradecanoylcarnitine, C5—isovalerylcarnitine, FN—false negatives, FP—false positives, His—histidine, LDA—linear discriminant analysis, LR—logistic regression, MeGlut—3-methylglutarylcarnitine, RR—ridge logistic regression, SVM—support vector machine, Trp—tryptophan.

Method	Features (Number)	Sensitivity % (FN)	Specificity %	FP
(a) Traditional NBS
NBS	C5	100	99.9951	103
(b) Full data set
LR	all (53)	100	99.9958	88
(c) Suspected diagnosis data set
LR	all (53)	98.2143 (1)	99.9981	41
RR	all (53)	96.2143 (2)	99.9987	29
SVM	all (51)	97.9286 (1)	99.9975	52
LR	Trp, C14OH, MeGlut, His, Asa	100	99.9978	47
RR	Trp, C14OH, MeGlut, His, Asa	100	99.9979	46
SVM	Trp, C14OH, MeGlut, His, Asa	100	99.9979	45
LDA-LR	LDA dimensions	100	99.9981	39
LDA-RR	LDA dimensions	100	99.9985	31
LDA-SVM	LDA dimensions	99.2857 (1)	99.9985	32