Fig 1: Dependency network of maternal autoantibody reactivity patterns.This network shows the top 70 patterns of autoantibody reactivity predictive of autism spectrum disorder. Each node represents a pattern. The closer two nodes are, the more similar are the sets of samples they cover. The bigger a node and the label, the more samples are covered by the corresponding pattern. Patterns identified by green nodes and bold fonts have 100% precision in the training as well as in the validation set; gray patterns are patterns that also have 100% precision in the training and the validation set, but are sub-patterns of at least one a green pattern, i.e., they cover a subset of samples of the green parent pattern (given this relation, they may be considered redundant); red patterns are patterns that are perfect with regard to precision in the training set but some samples fail in the validation set. Yellow patterns are also 100% precise in the training set but are absent in the validation set. Orange connections mean that the set of samples the bigger of one of the connected patterns covers is a subset of the set of samples the smaller connected pattern covers, while blue connections mean that the description of one pattern is a subset of another pattern (note that blue implies orange). CRMP1 and CRMP2 collapsin response mediator 1 and 2, GDA guanine deaminase, NSE neuron-specific enolase, LDHA-B lactate dehydrogenase A and B, STIP1 stress-induced phosphoprotein 1, and YBOX Y-box binding protein 1. We see that ASD is identified by three major patterns namely “CRMP1 = 1 AND CRMP2 = 1”, “STP = 1 AND NSE = 1” as well as “CRMP = 1 AND GDA = 1”. Most of the other patterns are sub-patterns or represent only a very small set of covered samples.