Differential expression (DE) analysis based on transcriptomic data provides a genome-wide assessment of gene responsiveness. We recently characterized transcriptional plasticity (TP)-the variability of gene expression in response to environmental changes-but its impact on DE analysis remained unexplored. In this work, we examined how TP affects DE analysis and introduced a TP-aware framework to improve the interpretation of DE results. We revealed correlations between fold change of gene expression and TP in 238 experiments with Mycobacterium tuberculosis (Mtb) and Escherichia coli (E. coli), which carried inherent biases, favoring genes with high TP while overlooking those with low TP. Therefore, we employed Locally Estimated Scatterplot Smoothing on TP to adjust the fold change of gene expression. Adjusted DE analyses identified new responsive pathways and yielded higher overall statistical significance and enrichment scores, especially for pathways with low-TP genes. Specifically, adjusted DE results revealed that bedaquiline treatment of Mtb induced cholesterol degradation, linezolid repressed acetate metabolism, and infection of macrophages upregulated fatty acid metabolism while downregulating cofactor biosynthesis. We also demonstrate that the adjustment strategy can be applied to other bacterial species and is compatible with various RNA-seq quantification approaches. In summary, we introduce a TP-aware approach that normalizes DE analysis by correcting for inherent transcriptional variability.

\n Contains fulltext :\n 237567.pdf (Publisher’s version ) (Open Access)\n

Abstract Motivation: Next-generation sequencing affords an efficient analysis of transposon insertion libraries, which can be used to identify essential genes in bacteria. To analyse this high-resolution data, we present a formal Bayesian framework for estimating the posterior probability of essentiality for each gene, using the extreme-value distribution to characterize the statistical significance of the longest region lacking insertions within a gene. We describe a sampling procedure based on the Metropolis–Hastings algorithm to calculate posterior probabilities of essentiality while simultaneously integrating over unknown internal parameters. Results: Using a sequence dataset from a transposon library for Mycobacterium tuberculosis, we show that this Bayesian approach predicts essential genes that correspond well with genes shown to be essential in previous studies. Furthermore, we show that by using the extreme-value distribution to characterize genomic regions lacking transposon insertions, this method is capable of identifying essential domains within genes. This approach can be used for analysing transposon libraries in other organisms and augmenting essentiality predictions with statistical confidence scores. Availability: A python script implementing the method described is available for download from http://saclab.tamu.edu/essentiality/. Contact: [email protected] or [email protected] Supplementary information: Supplementary data are available at Bioinformatics online.