Fig. 1.

VNN. The first layer of the VNN incorporates gene-level features, including gene mutations, copy number amplifications (CNAs), and copy number deletions (CNDs). Subsequent assembly layers aggregate gene-level features into assembly-level information, guided by the hierarchical relationships defined by a map of protein assemblies. The output state of each gene (g) and assembly (O) is represented by artificial neurons (1 neuron per gene, multiple neurons per assembly). Each node in the hierarchy indicates a protein assembly. An example path of information flow is shown in red. Adapted from Park et al. (29). RS-VNN, random set VNN.