Fig. 5.

DIABLO uncovered biologically relevant features by integrating information across data types. Schematic representation of two contrasting integration approaches using multivariate techniques: a shows that DIABLO selects features jointly across data types, resulting in the identification of features with strong associations across data types. Conversely, as shown in b, ensembles of multivariate models, constructed independently of each other, result in a selection of features that are poorly associated across data types. This is visualized in correlation heatmaps of the selected features and corresponding networks, with dense subgraphs, or network modules, encircled. In particular, the network modules identified in (a) include a number of features selected from all data types. This is not the case in b. The minimal set of features selected by DIABLO across data types as shown in c could discriminate between DOL and distinct sets of these features separated DOL0 from all other DOLs (DIABLO component 1) and DOL1, 3, and 7 from each other (DIABLO component 2). Features identified by DIABLO (blue bars) were largely distinct from those identified by more traditional single-OMICs multivariate approaches (red bars; overlaps in gray); shown in d using an UpSet plot. Moreover, features identified by DIABLO were more strongly enriched for known biological (functional) pathways; shown in e using an UpSet plot (blue vs. red bars). Horizontal bars are mapped to the number of elements in each set of features being compared. Vertical bars correspond to the number of elements in the intersections when carrying out various set comparisons. DIABLO: Data Integration Analysis for Biomarker discovery using Latent cOmponents, DOL: day of life