Modelling gene networks by non-linear analysis of microarray data.

The biochemical systems that build cells and organisms and keep them working depend on genetic control. The genome is the full set of controlling genes. Traditionally, scientists study individual genes in isolation because it has been too complicated to study them all simultaneously, but genes actually work together as part of interacting networks. Gene activity is controlled by transcription factors / proteins that switch genes on and off. The correct activation of the appropriate genes at the right time is essential for cells to work properly and to combat environmentally induced problems like damaged DNA. Failure to activate the correct combination of genes after DNA damage can lead to cancer. It is now possible to simultaneously measure how active all the genes in an organism are at a particular time using microarrays. Connecting these snapshots of gene activity gives a dynamic picture of how genes respond to cell stress and other signals. Intervention into gene networks could offer the opportunity to modify the response to bring about a better outcome / for example using drugs to change gene activity to enhance the cancer cell sensitivity to a chemotherapy agent. However, at present we are unable to predict how a network will respond to intervention. To make predictions about gene network activity it is not enough to simply know how much of each gene is there. It is also necessary to calculate the activity of the controlling transcription factor, to consider the sensitivity of a target gene to the transcription factor, and to estimate how much of it was there to start with, and how fast it degrades. We have developed a simple mathematical procedure (called HVDM) that combines these parameters to successfully model the network controlled by a single transcription factor (p53). In this new project we wish to develop a new mathematical tool that can identify all the main activities controlling transcription in the DNA damage response using only microarray data, and then model how these factors interact to produce the outcome observed. The finished product will be a computer tool which accurately predicts what will happen to the gene network in any number of possible scenarios. For example, predictions will be made about how the network would respond to a drug which affects one or more transcription factors, and what the effect of this is likely to be on the cells. The advantage of this approach is that it is more widely applicable and more accurate than other ways of analyzing microarrays. It can predict network activity using small datasets where experimental methods would require an impractical number of observations. The implications of this are that researchers will be able to predict gene networks activity much more accurately and efficiently than before. This will make more efficient use of very expensive research resources, and lead to better informed decisions when evaluating potential drug targets for clinical application

Time course microarray data contains hidden information about transcription factor activity and the sensitivity of target genes. Extraction of this information would allow the quantitative reconstruction of the network in silico, permitting predictions to be made about network behaviour. In this project, we will develop tools based on dynamical equation models which will extract the main activating forces in a complex gene response network. Models will be based on experimental data obtained after activating the DNA damage response network in MOLT4 and B-CLL-B human cell lines using irradiation. Previously we developed a simple linear model, Hidden Variable Dynamic Modelling, which was able to accurately predict targets of a single transcription factor, p53. We now plan to develop new models based on non-linear terms that will describe promoter saturation and threshold effects, co-regulation of transcripts by multiple transcription factors and will take account of enhancer effects, additive regulation and synergism. The new models will be ordinary differential equations with increasingly complex production terms. We will combine these non-linear models with a refined procedure for extracting multiple transcriptional activities from array data to simultaneously predict the behaviour of multiple transcription factors within the system. Each modelling stage will be followed by predictions of system behaviour at either different doses of irradiation, or with key regulators targeted. Predictions will be tested by independent experiments using gene knock-down techniques in MOLT4 and B-CLL-B lines. Discrepancies between model and data will drive model improvement. The end product will be a set of complex but user-friendly tools designed to efficiently and quantitatively analyse time course microarray data and to dissect out the relative contributions of different transcription factors to a biological response.