Gene expression studies often use bulk RNA sequencing of mixed cell populations because single cell or sorted cell sequencing may be prohibitively expensive. However, mixed cell studies may miss expression patterns that are restricted to specific cell populations. Computational deconvolution can be used to estimate cell fractions from bulk expression data and infer average cell-type expression in a set of samples (e.g., cases or controls), but imputing sample-level cell-type expression is required for more detailed analyses, such as relating expression to quantitative traits, and is less commonly addressed. Here, we assessed the accuracy of imputing sample-level cell-type expression using a real dataset where mixed peripheral blood mononuclear cells (PBMC) and sorted (CD4, CD8, CD14, CD19) RNA sequencing data were generated from the same subjects (N=158), and pseudobulk datasets synthesised from eQTLgen single cell RNA-seq data. We compared three domain-specific methods, CIBERSORTx, bMIND and debCAM/swCAM, and two cross-domain machine learning methods, multiple response LASSO and ridge, that had not been used for this task before. We also assessed the methods according to their ability to recover differential gene expression (DGE) results. LASSO/ridge showed higher sensitivity but lower specificity for recovering DGE signals seen in observed data compared to deconvolution methods, although LASSO/ridge had higher area under curves than deconvolution methods. Machine learning methods have the potential to outperform domain-specific methods when suitable training data are available.

Clustering is widely used in bioinformatics and many other fields, with applications from exploratory analysis to prediction. Many types of data have associated uncertainty or measurement error, but this is rarely used to inform the clustering. We present Dirichlet Process Mixtures with Uncertainty (DPMUnc), an extension of a Bayesian nonparametric clustering algorithm which makes use of the uncertainty associated with data points. We show that DPMUnc out-performs existing methods on simulated data. We cluster immune-mediated diseases (IMD) using GWAS summary statistics, which have uncertainty linked with the sample size of the study. DPMUnc separates autoimmune from autoinflammatory diseases and isolates other subgroups such as adult-onset arthritis. We additionally consider how DPMUnc can be used to cluster gene expression datasets that have been summarised using gene signatures. We first introduce a novel procedure for generating a summary of a gene signature on a dataset different to the one where it was discovered, which incorporates a measure of the variability in expression across signature genes within each individual. We summarise three public gene expression datasets containing patients with a range of IMD, using three relevant gene signatures. We find association between disease and the clusters returned by DPMUnc, with clustering structure replicated across the datasets. The significance of this work is two-fold. Firstly, we demonstrate that when data has associated uncertainty, this uncertainty should be used to inform clustering and we present a method which does this, DPMUnc. Secondly, we present a procedure for using gene signatures in datasets other than where they were originally defined. We show the value of this procedure by summarising gene expression data from patients with immune-mediated diseases using relevant gene signatures, and clustering these patients using DPMUnc.

MOTIVATION: While many pipelines have been developed for calling genotypes using RNA-sequencing (RNA-Seq) data, they all have adapted DNA genotype callers that do not model biases specific to RNA-Seq such as allele-specific expression (ASE). RESULTS: Here, we present Bayesian beta-binomial mixture model (BBmix), a Bayesian beta-binomial mixture model that first learns the expected distribution of read counts for each genotype, and then deploys those learned parameters to call genotypes probabilistically. We benchmarked our model on a wide variety of datasets and showed that our method generally performed better than competitors, mainly due to an increase of up to 1.4% in the accuracy of heterozygous calls, which may have a big impact in reducing false positive rate in applications sensitive to genotyping error such as ASE. Moreover, BBmix can be easily incorporated into standard pipelines for calling genotypes. We further show that parameters are generally transferable within datasets, such that a single learning run of less than 1 h is sufficient to call genotypes in a large number of samples. AVAILABILITY AND IMPLEMENTATION: We implemented BBmix as an R package that is available for free under a GPL-2 licence at https://gitlab.com/evigorito/bbmix and https://cran.r-project.org/package=bbmix with accompanying pipeline at https://gitlab.com/evigorito/bbmix_pipeline.

BACKGROUND: Cluster analysis is an integral part of precision medicine and systems biology, used to define groups of patients or biomolecules. Consensus clustering is an ensemble approach that is widely used in these areas, which combines the output from multiple runs of a non-deterministic clustering algorithm. Here we consider the application of consensus clustering to a broad class of heuristic clustering algorithms that can be derived from Bayesian mixture models (and extensions thereof) by adopting an early stopping criterion when performing sampling-based inference for these models. While the resulting approach is non-Bayesian, it inherits the usual benefits of consensus clustering, particularly in terms of computational scalability and providing assessments of clustering stability/robustness. RESULTS: In simulation studies, we show that our approach can successfully uncover the target clustering structure, while also exploring different plausible clusterings of the data. We show that, when a parallel computation environment is available, our approach offers significant reductions in runtime compared to performing sampling-based Bayesian inference for the underlying model, while retaining many of the practical benefits of the Bayesian approach, such as exploring different numbers of clusters. We propose a heuristic to decide upon ensemble size and the early stopping criterion, and then apply consensus clustering to a clustering algorithm derived from a Bayesian integrative clustering method. We use the resulting approach to perform an integrative analysis of three 'omics datasets for budding yeast and find clusters of co-expressed genes with shared regulatory proteins. We validate these clusters using data external to the analysis. CONCLUSTIONS: Our approach can be used as a wrapper for essentially any existing sampling-based Bayesian clustering implementation, and enables meaningful clustering analyses to be performed using such implementations, even when computational Bayesian inference is not feasible, e.g. due to poor exploration of the target density (often as a result of increasing numbers of features) or a limited computational budget that does not along sufficient samples to drawn from a single chain. This enables researchers to straightforwardly extend the applicability of existing software to much larger datasets, including implementations of sophisticated models such as those that jointly model multiple datasets.

BACKGROUND: Genome-wide association studies (GWAS) are limited in power to detect associations that exceed the stringent genome-wide significance threshold. This limitation can be alleviated by leveraging relevant auxiliary data, such as functional genomic data. Frameworks utilising the conditional false discovery rate have been developed for this purpose, and have been shown to increase power for GWAS discovery whilst controlling the false discovery rate. However, the methods are currently only applicable for continuous auxiliary data and cannot be used to leverage auxiliary data with a binary representation, such as whether SNPs are synonymous or non-synonymous, or whether they reside in regions of the genome with specific activity states. RESULTS: We describe an extension to the cFDR framework for binary auxiliary data, called "Binary cFDR". We demonstrate FDR control of our method using detailed simulations, and show that Binary cFDR performs better than a comparator method in terms of sensitivity and FDR control. We introduce an all-encompassing user-oriented CRAN R package ( https://annahutch.github.io/fcfdr/ ; https://cran.r-project.org/web/packages/fcfdr/index.html ) and demonstrate its utility in an application to type 1 diabetes, where we identify additional genetic associations. CONCLUSIONS: Our all-encompassing R package, fcfdr, serves as a comprehensive toolkit to unite GWAS and functional genomic data in order to increase statistical power to detect genetic associations.

MOTIVATION: Gene clustering and sample clustering are commonly used to find patterns in gene expression datasets. However, genes may cluster differently in heterogeneous samples (e.g. different tissues or disease states), whilst traditional methods assume that clusters are consistent across samples. Biclustering algorithms aim to solve this issue by performing sample clustering and gene clustering simultaneously. Existing reviews of biclustering algorithms have yet to include a number of more recent algorithms and have based comparisons on simplistic simulated datasets without specific evaluation of biclusters in real datasets, using less robust metrics. RESULTS: We compared four classes of sparse biclustering algorithms on a range of simulated and real datasets. All algorithms generally struggled on simulated datasets with a large number of genes or implanted biclusters. We found that Bayesian algorithms with strict sparsity constraints had high accuracy on the simulated datasets and did not require any post-processing, but were considerably slower than other algorithm classes. We found that non-negative matrix factorisation algorithms performed poorly, but could be re-purposed for biclustering through a sparsity-inducing post-processing procedure we introduce; one such algorithm was one of the most highly ranked on real datasets. In a multi-tissue knockout mouse RNA-seq dataset, the algorithms rarely returned clusters containing samples from multiple different tissues, whilst such clusters were identified in a human dataset of more closely related cell types (sorted blood cell subsets). This highlights the need for further thought in the design and analysis of multi-tissue studies to avoid differences between tissues dominating the analysis. AVAILABILITY: Code to run the analysis is available at https://github.com/nichollskc/biclust_comp, including wrappers for each algorithm, implementations of evaluation metrics, and code to simulate datasets and perform pre- and post-processing. The full tables of results are available at https://doi.org/10.5281/zenodo.4581206.

MOTIVATION: Polygenic scores (PGS) aim to genetically predict complex traits at an individual level. PGS are typically trained on genome-wide association summary statistics and require an independent test dataset to tune parameters. More recent methods allow parameters to be tuned on the training data, removing the need for independent test data, but approaches are computationally intensive. Based on fine-mapping principles, we present RápidoPGS, a flexible and fast method to compute PGS requiring summary-level Genome-wide association studies (GWAS) datasets only, with little computational requirements and no test data required for parameter tuning. RESULTS: We show that RápidoPGS performs slightly less well than two out of three other widely used PGS methods (LDpred2, PRScs and SBayesR) for case-control datasets, with median r2 difference: -0.0092, -0.0042 and 0.0064, respectively, but up to 17 000-fold faster with reduced computational requirements. RápidoPGS is implemented in R and can work with user-supplied summary statistics or download them from the GWAS catalog. AVAILABILITY AND IMPLEMENTATION: Our method is available with a GPL license as an R package from CRAN and GitHub. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Genome Wide Association Studies (GWAS) have successfully identified thousands of loci associated with human diseases. Bayesian genetic fine-mapping studies aim to identify the specific causal variants within GWAS loci responsible for each association, reporting credible sets of plausible causal variants, which are interpreted as containing the causal variant with some "coverage probability". Here, we use simulations to demonstrate that the coverage probabilities are over-conservative in most fine-mapping situations. We show that this is because fine-mapping data sets are not randomly selected from amongst all causal variants, but from amongst causal variants with larger effect sizes. We present a method to re-estimate the coverage of credible sets using rapid simulations based on the observed, or estimated, SNP correlation structure, we call this the "adjusted coverage estimate". This is extended to find "adjusted credible sets", which are the smallest set of variants such that their adjusted coverage estimate meets the target coverage. We use our method to improve the resolution of a fine-mapping study of type 1 diabetes. We found that in 27 out of 39 associated genomic regions our method could reduce the number of potentially causal variants to consider for follow-up, and found that none of the 95% or 99% credible sets required the inclusion of more variants-a pattern matched in simulations of well powered GWAS. Crucially, our method requires only GWAS summary statistics and remains accurate when SNP correlations are estimated from a large reference panel. Using our method to improve the resolution of fine-mapping studies will enable more efficient expenditure of resources in the follow-up process of annotating the variants in the credible set to determine the implicated genes and pathways in human diseases.

MOTIVATION: Methods for analysis of GWAS summary statistics have encouraged data sharing and democratized the analysis of different diseases. Ideal validation for such methods is application to simulated data, where some 'truth' is known. As GWAS increase in size, so does the computational complexity of such evaluations; standard practice repeatedly simulates and analyses genotype data for all individuals in an example study. RESULTS: We have developed a novel method based on an alternative approach, directly simulating GWAS summary data, without individual data as an intermediate step. We mathematically derive the expected statistics for any set of causal variants and their effect sizes, conditional upon control haplotype frequencies (available from public reference datasets). Simulation of GWAS summary output can be conducted independently of sample size by simulating random variates about these expected values. Across a range of scenarios, our method, produces very similar output to that from simulating individual genotypes with a substantial gain in speed even for modest sample sizes. Fast simulation of GWAS summary statistics will enable more complete and rapid evaluation of summary statistic methods as well as opening new potential avenues of research in fine mapping and gene set enrichment analysis. AVAILABILITY AND IMPLEMENTATION: Our method is available under a GPL license as an R package from http://github.com/chr1swallace/simGWAS. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Genome-wide association studies (GWAS) have identified many loci implicated in disease susceptibility. Integration of GWAS summary statistics (P-values) and functional genomic datasets should help to elucidate mechanisms. RESULTS: We extended a non-parametric SNP set enrichment method to test for enrichment of GWAS signals in functionally defined loci to a situation where only GWAS P-values are available. The approach is implemented in VSEAMS, a freely available software pipeline. We use VSEAMS to identify enrichment of type 1 diabetes (T1D) GWAS associations near genes that are targets for the transcription factors IKZF3, BATF and ESRRA. IKZF3 lies in a known T1D susceptibility region, while BATF and ESRRA overlap other immune disease susceptibility regions, validating our approach and suggesting novel avenues of research for T1D. AVAILABILITY AND IMPLEMENTATION: VSEAMS is available for download (http://github.com/ollyburren/vseams).

MOTIVATION: The invention of next-generation sequencing technology has made it possible to study the rare variants that are more likely to pinpoint causal disease genes. To make such experiments financially viable, DNA samples from several subjects are often pooled before sequencing. This induces large between-pool variation which, together with other sources of experimental error, creates over-dispersed data. Statistical analysis of pooled sequencing data needs to appropriately model this additional variance to avoid inflating the false-positive rate. RESULTS: We propose a new statistical method based on an extra-binomial model to address the over-dispersion and apply it to pooled case-control data. We demonstrate that our model provides a better fit to the data than either a standard binomial model or a traditional extra-binomial model proposed by Williams and can analyse both rare and common variants with lower or more variable pool depths compared to the other methods. AVAILABILITY: Package 'extraBinomial' is on http://cran.r-project.org/. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics Online.