Figure 1.
Block diagram of the sparsity-aware maximum likelihood (SML) algorithm.
The first and third blocks perform cross-validation to select optimal parameters and
to be used in (3) and (4), respectively. The second block produces weights
and error-variance estimate
after solving (4). Finally, the fourth block takes data
and
together with
,
and
and solves (3) to yield
, which represents the SML estimator for
in (1) revealing the genetic-interaction network. A more detailed description of the SML algorithm is given in Algorithm 1 in the Methods section.
Table Algorithm 1.
Algorithm 1. SML
Figure 2.
Performance of SML, AL and QDG algorithms for directed acyclic networks of [(a) and (b)] or 30 [(c) and (d)] genes.
Expected number of nodes per node is . PD and FDR were obtained from 100 replicates of the network with different sample sizes (
to 1,000).
Figure 3.
Performance of SML, AL and QDG algorithms for directed cyclic networks of [(a) and (b)] or 30 [(c) and (d)] genes.
Expected number of nodes per node is . PD and FDR were obtained from 100 replicates of the network with different sample sizes (
to 1,000).
Table 1.
Performance of SML, AL and QDG algorithms.
Figure 4.
Performance of the SML and AL algorithms for directed acyclic networks of genes [(a) power of detection, and (b) false discover rate].
Expected number of nodes per node is . PD and FDR were obtained from 10 replicates of the network with different sample sizes (
to 1,000).
Figure 5.
Performance of the SML algorithms for DAGs [(a) and (b)] or DCGs [(c) and (d)] of genes with an expected number of nodes per node
and error variance
.
PD and FDR were obtained from 100 replicates of the network with different sample sizes ( to 1,000).
Figure 6.
The network of 39 human genes inferred from gene expression and eQTL data with the SML algorithm.
The 39 genes related to the immune function were chosen from [45] to have a reliable eQTL per gene. The SML algorithm was run with stability selection and edges were detected at an . See Table 3 for the IDs and description of 39 genes. IGH in this figure corresponds to gene ID ENSG00000211897. A
edge stands for inhibitory effect and a
edge stands for activating effect.