An integrated machine learning approach for predicting DosR-regulated genes in Mycobacterium tuberculosis

Abstract

BackgroundDosR is an important regulator of the response to stress such as limited oxygen availability in Mycobacterium tuberculosis. Time course gene expression data enable us to dissect this response on the gene regulatory level. The mRNA expression profile of a regulator, however, is not necessarily a direct reflection of its activity. Knowing the transcription factor activity (TFA) can be exploited to predict novel target genes regulated by the same transcription factor. Various approaches have been proposed to reconstruct TFAs from gene expression data. Most of them capture only a first-order approximation to the complex transcriptional processes by assuming linear gene responses and linear dynamics in TFA, or ignore the temporal information in data from such systems.ResultsIn this paper, we approach the problem of inferring dynamic hidden TFAs using Gaussian processes (GP). We are able to model dynamic TFAs and to account for both linear and nonlinear gene responses. To test the validity of the proposed approach, we reconstruct the hidden TFA of p53, a tumour suppressor activated by DNA damage, using published time course gene expression data. Our reconstructed TFA is closer to the experimentally determined profile of p53 concentration than that from the original study. We then apply the model to time course gene expression data obtained from chemostat cultures of M. tuberculosis under reduced oxygen availability. After estimation of the TFA of DosR based on a number of known target genes using the GP model, we predict novel DosR-regulated genes: the parameters of the model are interpreted as relevance parameters indicating an existing functional relationship between TFA and gene expression. We further improve the prediction by integrating promoter sequence information in a logistic regression model. Apart from the documented DosR-regulated genes, our prediction yields ten novel genes under direct control of DosR.ConclusionsChemostat cultures are an ideal experimental system for controlling noise and variability when monitoring the response of bacterial organisms such as M. tuberculosis to finely controlled changes in culture conditions and available metabolites. Nonlinear hidden TFA dynamics of regulators can be reconstructed remarkably well with Gaussian processes from such data. Moreover, estimated parameters of the GP can be used to assess whether a gene is controlled by the reconstructed TFA or not. It is straightforward to combine these parameters with further information, such as the presence of binding motifs, to increase prediction accuracy.

Predictive network modeling of the high-resolution dynamic plant transcriptome in response to nitrate

“…As a first example of complex SSMs, Zhang et al . used gaussian processes dynamical models with nonlinear dynamics to infer the profile of a single transcription factor (the tumor suppressor p53) and explained the activity of a large collection of genes using that transcription factor only (without any other transcription factor-gene interaction) [ 39 ]. Another example is the linear dynamical system, which Beal et al .…”

“…Because our noise reduction state-space modeling algorithm is efficient, simple and tractable, as explained in the Materials and methods section, it can handle larger numbers of genes (we focused on 76 genes) than other SSM approaches, given enough genes [ 37 - 39 ].…”

Determining PTEN Functional Status by Network Component Deduced Transcription Factor Activities

“…However, expression levels of these TFs will not always be sufficient to reflect their activity since the activity of a transcription factor (TFA) is controlled by various post-translational modifications as well as co-activator and co-repressor activities. Previous works by us and others have shown that TFA can be best inferred from the transcript levels of its direct target genes, rather than its mRNA level using Network Component Analysis (NCA) [15] , [16] , [17] . NCA is a model-based decomposition method to deduce transcription factor activity (TFA) and regulation control strength (CS) of TFs from target gene expression and information of TF - gene interactions.…”

Adaptation to Environmental Stimuli within the Host: Two-Component Signal Transduction Systems of Mycobacterium tuberculosis

“…In particular, DosR recognizes a fairly conserved 18/20-bp palindromic sequence that is present with some variation upstream of the transcriptional units for these genes (Table 3) (4,25,26,42,55,116,125,162). Similar sequences are also found upstream of other genes outside this regulon, although their regulation by DosR remains unclear (116,185). Promoter regions from DosR regulon genes often contain two or more sets of recognition sequences, and cooperative binding by DosR to these sites is necessary for their full induction (24)(25)(26).…”