Table 1: Performance comparison of PathIntegrate Multi-View using pathways versus using the molecular-level COPDgene dataset (mean AUC and 95% CI, as well as the number of latent variables (LV) used).

In both pathway and molecular-level scenarios the model was used to predict binary COPD status. The molecular-level model was fit both with all molecules available in the datasets, as well as only those mapping to pathways. AUC values are averaged across 5-times repeated 5-fold cross validation.

	All omics	Metabolomics and proteomics	Metabolomics and transcriptomics	Transcriptomics and proteomics	Metabolomics	Proteomics	Transcriptomics
AUC (pathway)	0.70 (0.67, 0.72) (4 LV)	0.67 (0.66, 0.69) (3 LV)	0.69 (0.67, 0.71) (3 LV)	0.68 (0.66, 0.70) (4 LV)	0.63 (0.61, 0.64) (1 LV)	0.67 (0.66, 0.68) (3 LV)	0.65 (0.63, 0.66) (3 LV)
AUC (molecular)	0.70 (0.69, 0.72) (6 LV)	0.71 (0.70, 0.72) (2 LV)	0.70 (0.68, 0.71) (6 LV)	0.71 (0.70, 0.73) (7 LV)	0.66 (0.65, 0.69) (2 LV)	0.72 (0.71, 0.74) (3 LV)	0.68 (0.66, 0.69) (5 LV)
AUC (molecular – only those mapping to pathways)	0.72 (0.70, 0.74) (7 LV)	0.72 (0.70, 0.74) (2 LV)	0.67 (0.66, 0.69) (6 LV)	0.70 (0.69, 0.72) (6 LV)	0.68 (0.67, 0.7) (2 LV)	0.71 (0.70, 0.73) (3 LV)	0.66 (0.64, 0.68) (7 LV)