Fig. 3From: Prediction of microbial communities for urban metagenomics using neural network approachThe architecture of our proposed model MetaMLAnn. Starting from the left, the input layer with k nodes will receive k features respectively. Then, every input node will connect to all blocks, where each block contains p hidden units. The shared block (Bshare) is connected to every output label, and every other individual blocks (B1…Bm) is connected to its corresponding output label. It is regularized by the evolutionary (phylogenetic) microbial similarity matrixBack to article page