Fig. 7.

Latent structure for supervised multimodal networks. From [33] and adapted from [38]. Assume two data types, $\mathcal{G}$ and $\mathcal{X}$. S describes the subset of features within $\mathcal{X}$ that relate to the disease outcome but not with any features within $\mathcal{G}$; H describes the subset of features within $\mathcal{G}$ that relate to the disease outcome but not with any features within $\mathcal{X}$; G and X describe the network where the subset of features within $\mathcal{G}$ relate to the subset of features within $\mathcal{X}$ that ultimately relate to the outcome; G^′ and X^′ denote the subset of features that relate to each other but not with the outcome; and finally each data type will have independent noise not related to each other or the outcome