Fig. 4

a Bayesian network of serum biomarkers and background information variables estimated by a Hill-climbing algorithm and averaged over 1000 bootstrap samples. A line between two variables indicates an association independent of all other variables in the network. Line thickness corresponds to association strength measured as proportion of times an association was present in 1000 bootstrap sample networks (a thicker line indicates a stronger association). Node size corresponds to the number of connections. The network was estimated using discrete data (with biomarkers divided into low/high groups, with cut-off defined by the median biomarker concentrations in the controls). b Association strengths to ASD risk for each biomarker measured as proportion of times an association was present in 1000 bootstrap sample networks