Table 3.

Cluster level aggregation improves common machine learning metrics

Metric	Cluster	Gene	Gene P-values	Old cluster	Old cluster P-values
MCC	0.14	0.06	<3 × 10−08	0.06	<2 × 10−12
F1	0.18	0.12	<2 × 10−05	0.14	<6 × 10−04
Area under curve	0.57	0.52	<5 × 10−08	0.53	<2 × 10−05

P-values are calculated by using a Mann–Whitney test comparing the 16 ‘Gene’-level predictions and 28 ‘Old Cluster’-level predictions with 64 ‘Cluster’-level predictions. ‘MCC’ refers to a Matthews Correlation Coefficient, ‘F1’ to the F1 score and ‘Area under curve’ refers to the area under the receiver operating characteristic (ROC) curve.