Only considering interactions between proteins that appeared in both compared networks, we observed that Jaccard distances decreased with increasing inference threshold (Fig 2A).
In particular, distances between SI and HC networks were not only smaller but also declined faster with increasing inference threshold compared with distances between SI and HTP networks.
Although mean shortest path lengths within connected components remained constant in HTP and HC networks, corresponding values of SI networks decreased with increasing inference threshold (supplementary Fig S4a,b online).
Before knowledge of the Sib1 genotype was used, p(Sib2 AA) would have been the Hardy-Weinberg frequency for major homozygotes, p. However, with the Sib1 genotype, p(Sib2 AA|Sib1 AA), the additional constraint increases the probability to p +pq+(q /4), increasing inference accuracy by pq+(q /4).
We show that the increased IBD inference accuracy gained using our method improves the performance in the task of gene mapping.
In both correlation scenarios, increasing heritability increases inference efficiency for genetic correlations but reduces efficiency for the estimation of heritabilities and environmental correlations.
As there is only one hyper-parameter associated with the extra knowledge if this hyper-parameter increases the inference method would use wrong information and certainly the result would be worse than without any extra knowledge.
To further increase the inference stringency and to down-select a smaller number of genes for future experimental validation, we applied the following criterion: the expression of a gene must show statistically significant difference between the treated and the control at both duplicate spots to be inferred as a differentially expressed gene.
As the amount of available data increases, inferences should be easier to make, and weaker correlations could be captured.
This reduced selection bias and potential confounding effects, and increased the causal inference in this observational study.
