Hidden Markov Induced Dynamic Bayesian Networks

Re-implementation of a Hidden Markov induced Dynamic Bayesian Network, proposed by Zhu & Wang, 2015, for inferring gene regulatory networks from time-series gene expression data.

Requirements

• GEOparse
• NumPy, SciPy, Pandas
• Matplotlib
• requirements.txt is provided

Dataset

Drosophila gene expression data collected by Arbeitman et. al., 2002 can be downloaded from Gene Expression Omnibus. Samples were sorted to form time-series observations for genes of interest (eve, grf/lmd, twi, mlc-c, mhc1, prm, actn, 140up, 128up, msp300), and then binarized following Zhao et. al., 2006.

Organization

• run_structural_EM.py: main file for calling Structural EM algorithm
• data_processing.py: process raw microarray data into binary timeseries
• hmdbn.py: define, save, & load HMDBNs
• baum_welch.py: calculate posterior distribution with forward-backward algorithm & helper functions
• probs_update.py: update initial, transition, & emission probabilities and calculate BWBIC score
• visualization.py: plot posterior distribution of HMDBNs for all genes of interest

Usage

To run Structural Expectation Maximization to fit the HMDBN on the Drosophila dataset, use:

python run_structural_EM.py