Fig. 2
From: GNE: a deep learning framework for gene network inference by aggregating biological information
Overview of Gene Network Embedding (GNE) Framework for gene interaction prediction. On the left,one-hot encoded representation of gene is encoded to dense vector \(\mathbf {v}_{i}^{(s)}\) of dimension d×1 which captures topological properties and expression vector of gene is transformed to \(\mathbf {v}_{i}^{(a)}\) of dimension d×1 which aggregates the attribute information (Step 1). Next, concatenation of two embedded vectors (creates vector with dimension 2d×1) allows to combine strength of both network structure and attribute modeling. Then, nonlinear transformation of concatenated vector enables GNE to capture complex statistical relationships between network structure and attribute information and learn better representations (Step 2). Finally, these learned representation of dimension d×1 is transformed into a probability vector of length M×1 in output layer, which contains the predictive probability of gene vi to all the genes in the network. Conditional probability p(vj|vi) on output layer indicates the likelihood that gene vj is connected with gene vi (Step 3)