Evaluating the consistency of gene sets used in the analysis of bacterial gene expression data

IMPROVED PERFORMANCE OF GENE SET ANALYSIS ON GENOME-WIDE TRANSCRIPTOMICS DATA WHEN USING GENE ACTIVITY STATE ESTIMATES

Gene set analysis methods continue to be a popular and powerful method of evaluating genome-wide transcriptomics data. These approach require a priori grouping of genes into biologically meaningful sets, and then conducting downstream analyses at the set (instead of gene) level of analysis. Gene set analysis methods have been shown to yield more powerful statistical conclusions than single-gene analyses due to both reduced multiple testing penalties and potentially larger observed effects due to the aggregation of effects across multiple genes in the set. Traditionally, gene set analysis methods have been applied directly to normalized, log-transformed, transcriptomics data. Recently, efforts have been made to transform transcriptomics data to scales yielding more biologically interpretable results. For example, recently proposed models transform log-transformed transcriptomics data to a confidence metric (ranging between 0 and 100%) that a gene is active (roughly speaking, that the gene product is part of an active cellular mechanism). In this manuscript, we demonstrate, on both real and simulated transcriptomics data, that tests for differential expression between sets of genes using are typically more powerful when using gene activity state estimates as opposed to log-transformed gene expression data. Our analysis suggests further exploration of techniques to transform transcriptomics data to meaningful quantities for improved downstream inference.

Computing and Applying Atomic Regulons to Understand Gene Expression and Regulation

Understanding gene function and regulation is essential for the interpretation, prediction, and ultimate design of cell responses to changes in the environment. An important step toward meeting the challenge of understanding gene function and regulation is the identification of sets of genes that are always co-expressed. These gene sets, Atomic Regulons (ARs), represent fundamental units of function within a cell and could be used to associate genes of unknown function with cellular processes and to enable rational genetic engineering of cellular systems. Here, we describe an approach for inferring ARs that leverages large-scale expression data sets, gene context, and functional relationships among genes. We computed ARs for Escherichia coli based on 907 gene expression experiments and compared our results with gene clusters produced by two prevalent data-driven methods: Hierarchical clustering and k-means clustering. We compared ARs and purely data-driven gene clusters to the curated set of regulatory interactions for E. coli found in RegulonDB, showing that ARs are more consistent with gold standard regulons than are data-driven gene clusters. We further examined the consistency of ARs and data-driven gene clusters in the context of gene interactions predicted by Context Likelihood of Relatedness (CLR) analysis, finding that the ARs show better agreement with CLR predicted interactions. We determined the impact of increasing amounts of expression data on AR construction and find that while more data improve ARs, it is not necessary to use the full set of gene expression experiments available for E. coli to produce high quality ARs. In order to explore the conservation of co-regulated gene sets across different organisms, we computed ARs for Shewanella oneidensis, Pseudomonas aeruginosa, Thermus thermophilus, and Staphylococcus aureus, each of which represents increasing degrees of phylogenetic distance from E. coli. Comparison of the organism-specific ARs showed that the consistency of AR gene membership correlates with phylogenetic distance, but there is clear variability in the regulatory networks of closely related organisms. As large scale expression data sets become increasingly common for model and non-model organisms, comparative analyses of atomic regulons will provide valuable insights into fundamental regulatory modules used across the bacterial domain.

A Bayesian Framework for the Classification of Microbial Gene Activity States

Numerous methods for classifying gene activity states based on gene expression data have been proposed for use in downstream applications, such as incorporating transcriptomics data into metabolic models in order to improve resulting flux predictions. These methods often attempt to classify gene activity for each gene in each experimental condition as belonging to one of two states: active (the gene product is part of an active cellular mechanism) or inactive (the cellular mechanism is not active). These existing methods of classifying gene activity states suffer from multiple limitations, including enforcing unrealistic constraints on the overall proportions of active and inactive genes, failing to leverage a priori knowledge of gene co-regulation, failing to account for differences between genes, and failing to provide statistically meaningful confidence estimates. We propose a flexible Bayesian approach to classifying gene activity states based on a Gaussian mixture model. The model integrates genome-wide transcriptomics data from multiple conditions and information about gene co-regulation to provide activity state confidence estimates for each gene in each condition. We compare the performance of our novel method to existing methods on both simulated data and real data from 907 E. coli gene expression arrays, as well as a comparison with experimentally measured flux values in 29 conditions, demonstrating that our method provides more consistent and accurate results than existing methods across a variety of metrics.

Novel gene sets improve set-level classification of prokaryotic gene expression data

Background

Set-level classification of gene expression data has received significant attention recently. In this setting, high-dimensional vectors of features corresponding to genes are converted into lower-dimensional vectors of features corresponding to biologically interpretable gene sets. The dimensionality reduction brings the promise of a decreased risk of overfitting, potentially resulting in improved accuracy of the learned classifiers. However, recent empirical research has not confirmed this expectation. Here we hypothesize that the reported unfavorable classification results in the set-level framework were due to the adoption of unsuitable gene sets defined typically on the basis of the Gene ontology and the KEGG database of metabolic networks. We explore an alternative approach to defining gene sets, based on regulatory interactions, which we expect to collect genes with more correlated expression. We hypothesize that such more correlated gene sets will enable to learn more accurate classifiers.

Methods

We define two families of gene sets using information on regulatory interactions, and evaluate them on phenotype-classification tasks using public prokaryotic gene expression data sets. From each of the two gene-set families, we first select the best-performing subtype. The two selected subtypes are then evaluated on independent (testing) data sets against state-of-the-art gene sets and against the conventional gene-level approach.

Results

The novel gene sets are indeed more correlated than the conventional ones, and lead to significantly more accurate classifiers. The novel gene sets are indeed more correlated than the conventional ones, and lead to significantly more accurate classifiers.

Conclusion

Novel gene sets defined on the basis of regulatory interactions improve set-level classification of gene expression data. The experimental scripts and other material needed to reproduce the experiments are available at http://ida.felk.cvut.cz/novelgenesets.tar.gz.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-015-0786-7) contains supplementary material, which is available to authorized users.

Cautions about the reliability of pairwise gene correlations based on expression data

BACKGROUND

Rapid growth in the availability of genome-wide transcript abundance levels through gene expression microarrays and RNAseq promises to provide deep biological insights into the complex, genome-wide transcriptional behavior of single-celled organisms. However, this promise has not yet been fully realized.

RESULTS

We find that computation of pairwise gene associations (correlation; mutual information) across a set of 2782 total genome-wide expression samples from six diverse bacteria produces unexpectedly large variation in estimates of pairwise gene association-regardless of the metric used, the organism under study, or the number and source of the samples. We pinpoint the cause to sampling bias. In particular, in repositories of expression data (e.g., Gene Expression Omnibus, GEO), many individual genes show small differences in absolute gene expression levels across the set of samples. We demonstrate that these small differences are due mainly to "noise" instead of "signal" attributable to environmental or genetic perturbations. We show that downstream analysis using gene expression levels of genes with small differences yields biased estimates of pairwise association.

CONCLUSIONS

We propose flagging genes with small differences in absolute, RMA-normalized, expression levels (e.g., standard deviation less than 0.5), as potentially yielding biased pairwise association metrics. This strategy has the potential to substantially improve the confidence in genome-wide conclusions about transcriptional behavior in bacterial organisms. Further work is needed to further refine strategies to identify genes with small difference in expression levels prior to computing gene-gene association metrics.

The SEED and the Rapid Annotation of microbial genomes using Subsystems Technology (RAST)

In 2004, the SEED (http://pubseed.theseed.org/) was created to provide consistent and accurate genome annotations across thousands of genomes and as a platform for discovering and developing de novo annotations. The SEED is a constantly updated integration of genomic data with a genome database, web front end, API and server scripts. It is used by many scientists for predicting gene functions and discovering new pathways. In addition to being a powerful database for bioinformatics research, the SEED also houses subsystems (collections of functionally related protein families) and their derived FIGfams (protein families), which represent the core of the RAST annotation engine (http://rast.nmpdr.org/). When a new genome is submitted to RAST, genes are called and their annotations are made by comparison to the FIGfam collection. If the genome is made public, it is then housed within the SEED and its proteins populate the FIGfam collection. This annotation cycle has proven to be a robust and scalable solution to the problem of annotating the exponentially increasing number of genomes. To date, >12 000 users worldwide have annotated >60 000 distinct genomes using RAST. Here we describe the interconnectedness of the SEED database and RAST, the RAST annotation pipeline and updates to both resources.

Incorporating prior knowledge to increase the power of genome-wide association studies

Typical methods of analyzing genome-wide single nucleotide variant (SNV) data in cases and controls involve testing each variant's genotypes separately for phenotype association, and then using a substantial multiple-testing penalty to minimize the rate of false positives. This approach, however, can result in low power for modestly associated SNVs. Furthermore, simply looking at the most associated SNVs may not directly yield biological insights about disease etiology. SNVset methods attempt to address both limitations of the traditional approach by testing biologically meaningful sets of SNVs (e.g., genes or pathways). The number of tests run in a SNVset analysis is typically much lower (hundreds or thousands instead of millions) than in a traditional analysis, so the false-positive rate is lower. Additionally, by testing SNVsets that are biologically meaningful finding a significant set may more quickly yield insights into disease etiology.In this chapter we summarize the short history of SNVset testing and provide an overview of the many recently proposed methods. Furthermore, we provide detailed step-by-step instructions on how to perform a SNVset analysis, including a substantial number of practical tips and questions that researchers should consider before undertaking a SNVset analysis. Lastly, we describe a companion R package (snvset) that implements recently proposed SNVset methods. While SNVset testing is a new approach, with many new methods still being developed and many open questions, the promise of the approach is worth serious consideration when considering analytic methods for GWAS.