Fig. 1

The PathCORE-T software analysis workflow. a A user applies a feature construction algorithm to a transcriptomic dataset of genes-by-samples. The features constructed must preserve the genes in the dataset and assign weights to each of these genes according to some distribution. b Inputs required to run the complete PathCORE-T analysis workflow. The constructed features are stored in a weight matrix. Based on how gene weights are distributed in the constructed features, a user defines thresholds to select the set of genes most indicative of each feature’s function—we refer to these user-defined thresholds as gene signature rules. Finally, a list of pathway definitions will be used to interpret the features and build a pathway co-occurrence network. c Methods in the PathCORE-T analysis workflow (indicated using purple font) can be employed independently of each other so long as the necessary input(s) are provided. The 2 examples we describe to demonstrate PathCORE-T software use the following inputs: (1) the weight matrix and gene signature rules for eADAGE (applied to the P. aeruginosa gene compendium) and KEGG pathways and (2) the weight matrix and gene signature rules for NMF (applied to the TCGA pan-cancer dataset) and PID pathways.