Metabolomics variable selection and classification in the presence of observations below the detection limit using an extension of ERp

Table 2 LOO XERp Results for TBM vs Healthy Controls

Variable	% Selected	Classification Accuracy %			Average threshold	Direction
Variable	% Selected	Control Group	Experimental Group	Overall	Average threshold	Direction
4-hydroxyphenylpyruvic acid	100	100	82	94	0	Up
non-annotated-1	100	100	76	92	0	Up
methylcitric acid	100	94	88	92	0.45	Up
quinolinic acid	100	87	94	90	2.38	Up
2-hydroxybutyric acid	100	77	94	83	0.31	Up
2-hydroxyvaleric acid	100	77	94	83	0	Up
phenylacetylglutamine	100	90	71	83	1.02	Up
3-hydroxyisovaleric acid	100	71	100	81	20.02	Up
1,4-dihydroxycyclohexane	100	71	88	77	0.23	Up
2-ketoglutaric acid	100	65	82	71	0.53	Up
uracil	100	61	88	71	0.69	Up
non-annotated-3	75	100	71	90	0	Up
hexanoic acid	67	65	94	75	0.35	Down
isocitric acid	33	52	94	67	1.67	Up
pyruvic acid	17	81	76	79	0.25	Up

The table displays an excerpt of the LOO XERp results. The table is sorted in descending order of the second column (“% Selected”), i.e. the percentage of times a variable (as listed in the first column under “Variable”) was selected out of the \( N={N}_0+{N}_1 \) LOO iterations. In addition, the table reports the specificity (accurate classification of control subjects) and sensitivity (accurate classification of experimental subjects) percentages in columns 3 and 4, as well as the overall classification accuracy in column 5. Columns 6 and 7 contain the threshold values averaged over all instances the variable was selected (“Average Threshold”) and the direction of the shift found to be significant (“Direction”)

ISSN: 1471-2105