AR1MA1 - VBEM method for reverse engineering GRNs from pseudo-time series data

This repository includes the scripts to perform Gene Regulatory Network (GRN) inference from time series or pseudo-time series data using a first-order autoregressive moving average (AR1MA1) model within a variational Bayesian Expectation-Maximization (VBEM) framework.

Cite

A Bayesian framework for the inference of gene regulatory networks from time and pseudo-time series data, Bioinformatics (2017), DOI: 10.1093/bioinformatics/btx605

Quick guide

1. Download/clone this repository.
2. Set up your input data as here, with genes as rows and samples (pseudo-temporally sorted) as columns. The table sould be in Comma Separated Values (CSV) format, with a header (sample IDs) and column names (gene or proteins IDs).
3. Open the main script ( INFERENCE.m ) in Matlab and run it ( F5 ).
4. Use the dialog box to pick up the input CSV file.
5. The command window will show the inference progress.
6. The results will be saved in a text file using Simple Interaction File (SIF) extended format, with same filename and suffix __AR1MA1_GRN_inference.txt. The output file includes a header and the next columns: (i) Parent node, (ii) interaction type ("-|" for inhibition and "->" for activation), (iii) child node, (iv) the interaction weight, (v) the posterior probability and (vi) a score computed as the product of the posterior probability and the weight (normalized to the maximum value).

Implementation

The method is implemented in MATLAB, version 8.6 (R2015b). For previous/posterior versions some of the commands might be updated.

The inference method

The AR1MA1-VBEM method is explained in detail in the 4th chapter of my PhD thesis:

Bayesian methods for the inference of GRNs and protein profiles from gene expression microarrays data, 2012, ISBN: 978-84-9028-501-5

An early VBEM based approach, based on a first-order autoregressive (AR1) model, was published in:

A Survey of Statistical Models for Reverse Engineering Gene Regulatory Networks, 2009, DOI: 10.1109/MSP.2008.930647

Example: mESC

We analysed single-cell qPCR expression data of 46 genes and 280 samples from mouse zygote to blastocyst as presented in:

Resolution of cell fate decisions revealed by single-cell gene expression analysis from zygote to blastocyst, 2010, Developmental Cell, PubMed: 20412781

Data processing

Data were downloaded from source, processed to select cells within the oocyte-to-epiblast stages and pseudo-temporal sorted according to the hierarchical optimal-leaf ordering algorithm. The data, already processed and with the required input format, can be found here.

Results

The results can be found here in SIF format. A plot with the network from these results can be found here. The network has been plotted with Cytoscape, using this visual style.