Premium
Dynamic Bayesian Graphical Modeling to Predict Regulatory Networks in Hypertensive Rats
Author(s) -
Dayton Alex,
Ahn Kwang Woo,
Liu Pengyuan,
Laud Purushottam,
Stingo Francesco,
Vannucci Marina,
Bukowy John D,
Liang Mingyu,
Cowley Allen W
Publication year - 2017
Publication title -
the faseb journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.709
H-Index - 277
eISSN - 1530-6860
pISSN - 0892-6638
DOI - 10.1096/fasebj.31.1_supplement.722.7
Subject(s) - graphical model , gene , computational biology , bayesian probability , bayesian network , series (stratigraphy) , computer science , bioinformatics , biology , algorithm , artificial intelligence , genetics , paleontology
We have previously used Bayesian graphical methods to detect networks of interacting genes relevant to the development of salt sensitive hypertension. However, these previous approaches are “static” or time‐insensitive methods that cannot infer causality to gene interactions. Here, we present a dynamic Bayesian graphical modeling approach which is used to connect potential target genes to other genes and networks. Dynamic Bayesian graphical networking approaches require large amounts of time‐series data, in addition to being computationally complex. Data was generated by isolating medullary thick ascending limb (mTAL) cells from 3 rat strains: the Dahl Salt‐sensitive rat (SS), the Fawn Hooded Hypertensive rat (FHH) and the Spontaneously Hypertensive Rat (SHR). After isolation, the freshly isolated mTAL cells were exposed to one of two stimuli: H 2 O 2 (500 nM) or TNF‐α (50 ng/mL). RNA was then collected at 30 minute intervals for 8 hours, resulting in a 16‐point time‐series for each strain and stimulus. Each time‐series was duplicated, and duplicates were pooled to reduce overall measurement noise. As there are 6 combinations of strains (3) and stimuli (2) this procedure resulted in 96 individual time‐points. RNA was then isolated and sequenced on an Illumina Hi‐Seq 2500. The Dynamic Bayesian graphical model attempts to find connections between genes by looking at the sequence of activation/inactivation of genes across the entire 8 hour time‐series. For example, in a time‐series generated by treatment of SS mTAL cells with 500 nM H 2 O 2 , we find that the gene Dusp5 undergoes a 4‐fold increase in RNA expression one hour before a similar 4‐fold increase in the gene Rasgrp3, a GWAS gene for hypertension that has no known function in the mTAL. The Bayesian graphical modeling algorithm looks at such changes across each time‐series for the different rat strains and stimuli, and detects the most probable connections between the analyzed genes. This analysis represents the first attempt to create a dynamic Bayesian graph in non‐cancer mammalian cells. The use of such a graphical modeling approach in mTAL, a cell type of particular relevance to the pathophysiology of hypertension, will give us insight into the regulation of critical genes in hypertension and allow us to better understand, and potentially treat, this important disease. Support or Funding Information A. Dayton supported by NIH Fellowship: 1F30HL127979‐01A1.Work funded by NIH grant: 4P01HL116264.