Determination of prokaryotic transcriptional regulatory networks

Given their comparatively simple genome organization and regulatory systems, it is likely that a prokaryotic organism will be the first cellular organism with a predictive mathematical model for mapping inputs (genetic and environmental perturbations) to outputs (changes in growth rate and gene expression). We developed, validated, and optimized algorithms and experimental technologies to construct the comprehensive transcriptional regulatory interaction skeleton necessary for such a model. First, we developed and validated machine learning algorithms for inferring genome scale transcriptional regulation from a compendium of 597 Escherichia coli Affymetrix microarrays, predicting over one-thousand regulatory interactions at a 60% true positive rate. To enable rapid in vivo experimental validation of these computationally inferred interactions or discovery of novel transcriptional regulatory interactions with no prior knowledge, we then used statistical experimental design techniques to increase the throughput of a standard Chromatin Immunoprecipitation protocol by ten-fold, while also improving the signal-to-noise ratio. Finally, we developed a novel experimental method, using error-correcting DNA-sequence barcodes combined with highly-parallel sequencing of mRNA across multiple conditions, to experimentally define a species' transcriptional units and enable network inference in a more natural multi-species environment.