We generate synthetic data (sample size 300) according to a DAG composed
by 100 nodes and 107 edges (exactly as in Figure 1).
Each phenotype node is affected by three
QTLs, and we allow only additive genetic effects. The QTLs for each phenotype are
randomly
selected among 200 markers, with 10 markers unevenly distributed on each of 20 autosomes.
We allowed different
phenotypes to potentially share common QTLs. For each phenotype,
the regression coefficients with other phenotypes are
chosen uniformly between 0.5 and 1;
QTL effects are chosen between 0.2 to 0.6; and residual standard deviations are chosen
from 0.1 to 0.5. For each realization we apply the QDG algorithm to infer causal
directions for the edges of the skeleton
obtained by the PC-skeleton algorithm.
Usage
data(acyclic)
Details
For cyclic graphs, the output of the qdg function computes the
log-likelihood
up to the normalization constant (un-normalized log-likelihood). We can
use the un-normalized
log-likelihood to compare cyclic graphs with reversed directions (since they
have the same normalization constant). However we cannot compare cyclic and
acyclic graphs.
References
Chaibub Neto et al. (2008) Inferring causal phenotype networks from
segregating populations. Genetics 179: 1089-1100.