Fig. 2

Newborn metabolic pattern analysis with Random Forest (RF). (a) Receiver operating characteristic (ROC) curve analysis for newborns with and without a confirmed methylmalonic acidemia (MMA) diagnosis using RF analysis of 46 MS/MS analytes. Without altering the 96.1% sensitivity of MMA screening based on the primary newborn screening (NBS) analyte C3 and C3/C2, RF reduced the number of MMA false-positive cases to 49% (vertical dotted line), which increased the positive predictive value (PPV) from 16.5 to 28.9%. (b) The mean decrease in accuracy (MDA) was used to measure the contribution of individual metabolic analytes in the RF model. The relative importance of analytes for MMA metabolic pattern recognition is ranked from top to bottom with primary markers C3 and C3/C2 in bold. AUC area under the curve.