Raw repfdr::repfdr() results from which the graphical state assignments were determined
Format
List:
repfdr_em_resa list with
repfdr's EM resultsrepfdr_clustersrepfdr's configurationsrepfdr_clusters_strrepfdr's configurations, string representationrepfdr_clusters_piconfiguration's inferred priors
Details
repfdr::repfdr() is an algorithm suggested by Yekutieli and Heller in 2014
(Bioinformatics) for analysis of p-values or z-scores from different resources.
It is based on the assumption that z-scores from each resource (a time point
from a specific sex in our case) is either positive, null, or negative.
Then the algorithm internally learns the mixture distribution in each resource
and the dependencies among them. In the process, a simplifying assumption is
made about conditional independence between the z-scores given their state
(so for example, two high scores from weeks 4 and 8 in males are independent
given that we know that they are positive, non-null cases).
Our analysis pipeline is as follows: we run repfdr::repfdr() on the z-score data matrix
of all training-regulated features at 5% IWH FDR (see REPFDR_INPUTS). Then we use the output to extract
the analytes of each state in each time point. Here a state means one of
(male up, male null, male down) x (female up, female null, female down)
for each time point. Once these are selected we call them the node_sets.
Then, for each pair of node (x,y) such that y is from a time point that is adjacent
and after x (e.g., x is a node from week 4 and y is a node from week 8),
we define their edge set as the intersection of their analytes.
Reproduce our analysis with MotrpacRatTraining6mo::bayesian_graphical_clustering()
and MotrpacRatTraining6mo::repfdr_wrapper().