Table 5 Comparison of model performance.

From: Mass spectrometry combined with machine learning identifies novel protein signatures as demonstrated with multisystem inflammatory syndrome in children

Signature

Proteins

References

Original AUC

AUC and 95% Confidence Interval

CV-corrected AUC (%)

1

ORM1, SERPINA3, AZGP1

Current manuscript

–

88.0% (77.4–98.7%)

79.7

2

VWF, FCGBP, SERPINA3

Current manuscript

–

95.6% (89.6–100%)

94.2

3

LCP1, SERPINA3, BCHE

Nygaard et al.16

100% (test set)

94% (internal validation)

87% (external validation)

95.8% (91.2–100%)

89.7

4

CD163, PCSK9, [CXCL9]*

Yeoh et al.17

85.7% (76.6–94.8%)

94.1% (87.8–100%)

82.4

  1. AUC area-under-the-curve, CV cross-validation.
  2. *Yeoh et al.17 included CXCL9 in their protein signature, but we excluded it since we have no data for that protein.
Back to article page