Log-linear Graphical Model
: inferring probabilistic conditional independency from combinatorial regulation of transcription factors
To run LLGM,
please log in OpenLooper and visit this page again.
Transcription factors (TFs) regulate gene expressions by intricate combinatorial interactions. Computational modeling to infer the interactions is essential to understand the nature of transcriptional regulation. Log-linear Graphical Model (LLGM) deals with spurious TF interactions that may lead to deceptive inferences. Given a discrete (0 or 1)
input data matrix of TF-DNA binding instances, the LLGM estimates the probabilistic conditional independency for detecting the spuriousness, which is known as "Simpson's Paradox"
The format of the input matrix file:
Input data is a n x m matrix, where n is the number of promoters (genes) and m column is the number of TFs.
The input file has to include a line "#ITEM: TF1 TF2 TF3", where "TF1" is an item.
The absence or presence of one item is denoted by "1 0" or "0 1".
Each line has to be TAB deliminated; e.g. 'Promo_A[tab]0[tab]1[tab]1[tab]0[tab]1[tab]0'.
'Promo_A[tab]0[tab]1[tab]1[tab]0[tab]1[tab]0' means that the "Promo_A" includes
[see an example input data]
Two thresholds are required for running:
- "0 1" (=presence) of TF1
- "1 0" (=absence) of TF2
- "1 0" (=absence) of TF3.
In this web-version, the matrix size n and m is restricted to 50≤ n ≤2000 and 2≤ m ≤10.
You can request it for an extension by contacting us.
- p-value cutoff for the test of deviance of a current RM (reduced model) from the FM (full model)
- p-value cutoff for the test of deviance of a current RM (reduced model) from the previous RM
1. Lauritzen, S.L., "Graphical Models", Oxford University Press, 1996
2. Christensen, R., "Log-Linear Models and Logistic Regression", Springer-Verlag, 1997
3. Park SJ, Umemoto T, Saito-Adachi M, Shiratsuchi Y, Yamato M, Nakai K, "Computational Promoter Modeling Identifies the Modes of Transcriptional Regulation in Hematopoietic Stem Cells", PLoS ONE 9(4): e93853, 2014