Figure 2 | Scientific Reports

Figure 2

From: Mixed-effect Bayesian network reveals personal effects of nutrition

Figure 2

In (a) a bipartite Bayesian network is constructed from a few of the variables in Sysdimet data. The overall network factorizes into subnetworks \(G_1\) and \(G_2\) that form independently estimated local probability distributions. In (b) a general concentration-specific graphical model \(G_i\) is shown. It includes p observed random variables X for the nutrient levels and the corresponding concentration level \(Y_i\). The latent variables of each nutritional effect are estimated from the data. Personal variations \({\mathrm {b_i}}\) are assumed to follow Normal distribution, but blood concentrations can also be better modeled with Gamma distribution. Evaluation of the directed graphical model starts from the prior and the observed nodes. Their sampled values are propagated as input to downward nodes, and finally, the linear predictor (3) in the concentration node \(Y_i\) gathers them all and results in an estimation.

Back to article page