Fig 3. Diagram of latent variables in a network system with the outcome y.

Each of the latent variables (S, H, X, G, X′, G′, and Noise) describe a subset of features within data types $\mathcal{G}$ and $\mathcal{X}$ with their corresponding relationships. S describes the set of features within $\mathcal{X}$ that are related to y but not any features in $\mathcal{G}$; H describes the set of features within $\mathcal{G}$ that are related to y but not any features in $\mathcal{X}$; X describes the network features within $\mathcal{X}$ which is related to the network features within $\mathcal{G}$ and y (note that there are $p\prime$ features in this subset); G describes the network features within $\mathcal{G}$ which is related to the network features within $\mathcal{X}$ and y (note that there are $q\prime$ features in this subset); X′ describes the set of features within $\mathcal{X}$ that are related to a set of features within $\mathcal{G}$ but not related to y; G′ describes the set of features within $\mathcal{G}$ that are related to a set of features within $\mathcal{X}$ but not related to y; and Noise describes the features within $\mathcal{X}$ and $\mathcal{G}$ that are not related to each other or y. The variables within the respective dotted boxes summarize the various components contained within each of the data types, and the arrows signify a relationship and imply correlation between the corresponding components. SuMO-Fil thus describes “irrelevant” features as those belonging to components S, H, X′, G′, and Noise. Note that this figure was adapted from [17].