Integrating genome-scale metabolic modelling and transfer learning for human gene regulatory network reconstruction

The raw data is contained in the folders human and mouse. In each folder there is a README file explaining the relative contents.

main.m creates the TRFBA model. It loads the COBRA model for the organism and the associated data (gene expression, gene regulators and targets, IDs and remaining parameters), which is already in a MATLAB structure.

The MILProblem resulting from main.m is used by Geneknockout_and_fluxes_generation.m to compute the metabolic fluxes following the gene knockouts.

The result, deleted the fluxes smaller than 1e-7 for all the gene knockouts, can be found in Fluxes data (fluxes_single_KO_final.csv)

Ranking computation and heatmap generation.ipynb uses this result and the feature ranking in FeatureRanking.xlsx to compute the final feature ranking at the flux level contained in final_ranking.csv and final_index.csv.

final_ranking.csv and final_index.csv are used in Average_weight_pathway_and_FEA.m to conduct the flux enrichment analysis and compute the mean weight for each pathway.

Files Generate_figures.R and Ranking computation and heatmap generation.ipynb are used to generate the figures presented in the paper. The latter generates the heatmap and prepares the data to be used in the former, which generates the Euler-Venn diagram and the histograms.

- The experiments for feature ranking and the results (feature_ranking.zip)