Figure 2

Prior influence on reconstruction performance of inhibMCMC. Diagnostics for inhibMCMC of a randomly sampled network (N = 15), 50000 iterations, burn-in 5000, γ = 1, varying λ. The sampled network was used as prior confidence, i.e. the prior knowledge was 'perfect' in this test. Left: The smaller λ, the stronger the prior influence was and the closer the inferred networks were to the prior (reflected in increasing AUCs). Right: Comparison of Likelihood and Prior ratios, depending on λ. λ should be chosen such that the prior and likelihood ratios vary in a comparable range. For instance, based on the plot, set λ = 0.005.