Application of new probabilistic graphical models in the genetic regulatory networks studies

TL;DR

This paper presents two novel probabilistic graphical models for reconstructing genetic regulatory networks from DNA microarray data, highlighting their differences and limitations in capturing complex gene interactions.

Contribution

Introduces two new probabilistic graphical models, Independence Graph and Gaussian Network, with novel search algorithms for genetic network reconstruction.

Findings

IG model yields sparse graphs

GN model produces dense, information-rich graphs

Sample size and data complexity limit prediction accuracy

Abstract

This paper introduces two new probabilistic graphical models for reconstruction of genetic regulatory networks using DNA microarray data. One is an Independence Graph (IG) model with either a forward or a backward search algorithm and the other one is a Gaussian Network (GN) model with a novel greedy search method. The performances of both models were evaluated on four MAPK pathways in yeast and three simulated data sets. Generally, an IG model provides a sparse graph but a GN model produces a dense graph where more information about gene-gene interactions is preserved. Additionally, we found two key limitations in the prediction of genetic regulatory networks using DNA microarray data, the first is the sufficiency of sample size and the second is the complexity of network structures may not be captured without additional data at the protein level. Those limitations are present in all prediction methods which used only DNA microarray data.