Inference of Transcription Factor Concentrations and Regulatory Intensities

This page describes examples of how to use the Dynamical Model for Variational Inference of Protein Concentration and Regulatory Intensities detailed in . The code is available for download here.

Release Information

Current release is 0.11.

Release 0.11 contains a bug fix for file chipVarEstepCMu.m

Release 0.1 is the first release associated with the paper, containing scripts for recreating the results given.

The toolboxes required to run the code are listed below.

Toolbox	Version
NDLUTIL	0.15

Finally you will also need the NETLAB toolbox in your path.

Results in the Paper

The main scripts are demTuVar and demSpellmanVar which run the code on the metabolic and cell cycle data sets. Also of interest may be the script demFakeVar which runs the code on artificial data. demTuVar and demSpellmanVar invoke the functions chipVarEMmu and chipVarEM respectively which compute the posterior estimates of the latent variables by maximising the variational lower bound on the likelihood. The functions used are different as there is no need to estimate the baseline expression level for cDNA array data such as the cell cycle data set. The scripts output a variable model, containing the parameter values which maximise the likelihood, and expectationsB (posterior expectations of the regulatory intensities) and expectationsC, posterior expectations of the protein concentration profiles. The function chipVarEMmu also outputs a variable expectationsMu containing posterior statistics for the distribution of the baseline expression levels. The variable expectationsC has four fields: .entropy, a number contatining the posterior entropy of the approximating distribution, .c, a matrix with as many rows as transcription factors and as many columns as time points, .ccT, a 3D array containing the correlations among transcription factors at each time point, and .cAltc, a 3D array containing correlations among transcription factors at consecutive time points. By specialising the transcription factor index one can obtain plots of the transcription factor protein concentrations with errorbars. An example is show below for the transcription factor ACE2 during the cell cycle.
Inferred protein concentration profile for the transcription factor ACE2 during the yeast cell cycle. Similarly, expectationsB has several fields, but only two of them are of interest for data analysis, .b and .errorbars (the others are necessary for the estimation of the variational lower bound). The expectationsB.b variable is a matrix containing in each row the regulatory intensity with which a gene is affected by each transcription factor. Obviously most of the entries will be zero due to the sparsity of the network structure. The matrix expectationsB.errorbars contains the associated error bars. By specifying to a transcription factor, we can obtain a distribution of the significance levels of its regulatory intensities. An example of this is shown below for the transcription factor ACE2 during the cell cycle.
Significance of the regulatory intensities for the transcription factor ACE2 during the yeast cell cycle. The plot shows the ratio between gene-specific regulatory intensities and the associated noise