Fig. 1

Building and using the ONN model for microbial source tracking. a The sample data representation and training process of ONN model. (i) Sample data are transformed into the Matrix. With the Matrix, each column represents a taxonomic level and each row represents a feature; (ii) In parallel, samples are mapped to biome ontology according to their niches; (iii) The model is built and updated according to both samples’ abundance matrices and biome ontology information. More details about building, testing, and using the ONN model for source tracking are illustrated in Supplementary Figs. 1 and 2. b An illustrated example of microbial source tracking procedure using ONN4MST. (i) The input is the community structure of a real microbial community sample (this sample is from the biome “Root-Host_associated-Human-Digestive_system-Oral-Saliva”) that has been preprocessed and the Matrix has been provided into the model; (ii) Source tracking process at different layers. The red arrows indicate the search process from layer 1 to layer 6, accompanied with source contribution annotated in red. To compare with the procedure of ONN4MST, the yellow and blue arrows indicated the source tracking results (among the overall top 5 sources) of FEAST and SourceTracker, together with their source contributions, respectively. The actual biome is annotated by a red checkmark; (iii) The predicted biomes (with source contributions) by ONN4MST, FEAST, and SourceTracker