Figure 2 | Scientific Reports

Figure 2

From: Predicting Causal Relationships from Biological Data: Applying Automated Causal Discovery on Mass Cytometry Data of Human Immune Cells

Figure 2

Causal discovery using conditional independence. (a–d) Joint and conditional distributions under conditional independence. Black/red points indicate no activation/activation with PMA. Black/red bars show the mean values. Conditional protein is discretized in 10 levels. (a,b) Stimulation with PMA significantly increases the levels of both pErk and pZap70. (c) pZap70 is independent of PMA stimulation given the levels of pErk: P(pZap70|PMA, pErk) = P(pZap70|pErk); mean values for red and black points are very similar. (d) In contrast, PMA stimulation is associated with pErk even after conditioning on pZap70. (e) Possible causal Bayesian networks for three variables that are pairwisely dependent, and one of them is restricted to be a source variable (no incoming edges): conditional independence can distinguish among the causal, independent and full model. In the full model, the line between variables S and T does not have any arrowheads; causal direction of this edge cannot be determined from the data.

Back to article page