Fig 4. Overview of the experimental and computational approach to identify secreted factors of HSCs regulating HCC gene expression.

Conditioned medium of primary human HSC (n = 15) was transferred onto human Hep3B HCC cells. Gene expression data of HSC and HCC cells were filtered to reduce the dimensionality of the data and to build cause-and-effect (target) matrices. These served as input for the IDA algorithm which estimates causal effects for each cause on each target gene. Causal effects that were stable across sub-sampling runs (i.e. that were stable with respect to small perturbations of the data) were retained and subjected to Model-based Gene Set Analysis (MGSA) to extract a sparse set of HSC genes influencing HCC cell gene expression.