BioC 2018: Where Software and Biology Connect

When: July 25 (Developer Day) 26 and 27, 2018
Where: Victoria University, University of Toronto, Toronto, Canada
Twitter: #bioc2018

Community-Contributed Talks (Tentative)

• See the Conference Schedule for invited talks.

• Lee S*, Cook D, Lawrence M. plyranges: a fluent interface to Bioconductor’s Ranges infrastructure. The Bioconductor project has created many powerful S4 classes for reasoning about genomics data, such as the SummarizedExperiment class for representing experimental assays and the Ranges family of classes for representing genomic intervals. For new users of Bioconductor who are unfamiliar with object oriented programming or genomic data analysis, the learning curve for these classes is steep. New users often find themselves asking: how do I get my data into an appropriate class for my analysis? What are accessors and why do I have to use them? Why does this function return an object of class that I have not seen before? This results in analysts trying to solve problems that have been efficiently dealt with in other Bioconductor packages or writing code they may not fully understand.

  The goal of a fluent interface is to enable users to write human-readable code via method chaining and consistent function returns. Fluent interfaces fit naturally in the context of Bioinformatics workflows because they enable writing succinct pipelines. We have developed plyranges a package that attempts to construct a fluent interface to the Ranges classes defined in the IRanges and GenomicRanges packages. This package is inspired by dplyr and implements and extends its grammar of data manipulation to Ranges. We have defined methods for constructing, grouping, mutating, filtering, and summarising Ranges. We have also defined an algebra for reasoning about actions on Ranges and relationships between Ranges. By having an expansive grammar, we hope to cover the majority of analysis tasks a new user may face and thereby enable users to write clearer and more reproducible code.

• Zhun Miao*, Ke Deng, Xiaowo Wang, Xuegong Zhang. DEsingle for detecting three types of differential expression in single-cell RNA-seq data. The excessive amount of zeros in single-cell RNA-seq (scRNA-seq) data includes ‘real’ zeros due to the on-off nature of gene transcription in single cells and ‘dropout’ zeros due to technical reasons. Existing differential expression (DE) analysis methods cannot distinguish these two types of zeros.

  We developed an R package DEsingle which employed Zero-Inflated Negative Binomial model to estimate the proportion of real and dropout zeros and to define and detect three types of DE genes in scRNA-seq data, with regard to different expression status (DEs), differential expression abundance (DEa), and general differential expression (DEg).

  Results showed that DEsingle outperforms existing methods for scRNA-seq DE analysis, and can reveal different types of DE genes that are enriched in different biological functions.

• Abbas Rizvi*, Ezgi Karaesmen*, Leah Preus, Michael Sovic, Junke Wang, Lara Sucheston-Campbell. gwasurvivr: an R package to perform survival association testing on imputed genetic data. Increasingly researchers have become interested in time-to-event outcomes in the context of genetic variation. Existing software for performing survival analyses across millions of SNPs are limited. Recently, comprehensive stand-alone software packages, genipe and SurvivalGWAS_SV, were developed, however, they require user interaction with the raw output after imputation opening room for error during analysis. GWASTools, while available in R and can implement survival, is primarily for storing large SNP datasets and rigorous QC/QA. To address this unmet need, we developed an R/Bioconductor package to conduct fast and efficient genome wide survival analyses for on imputed genetic data generated using IMPUTE2 and VCF data generated from Michigan or Sanger imputation servers.

  gwasurvivr implements Cox proportional hazards models to test SNP association with outcome; the package allows for covariates and SNP-covariate interaction. To potentially decrease the number of iterations needed for convergence when optimizing the parameters estimates in the Cox model, we modified survival::coxph.fit, to fit covariates without the SNP and use those parameter estimates as initial starting points. For models without covariates the parameter estimation optimization begins with null initial value. Users can internally subset the data by providing sample IDs and pre-filter SNPs by info score and MAF. Output for each SNP includes parameter estimates, p-values, MAFs, INFO scores, number of events and total sample N. gwasurvivr is well-suited for multi-core processors and users can specify node preferences used during computation. To overcome R memory limitations gwasurvivr iteratively performs survival on subsets of the entire data.

  We benchmarked our package with genipe, SurvivalGWAS_SV, and GWASTools using IMPUTE2 data for varying sample sizes (n=100, n=1000, n=5000) and SNPs (p=1000, p=10000, p=100000) including two non-genetic covariates. All packages showed excellent agreement across MAF estimates, coefficient estimates, and p-values, with gwasurvivr outperforming the other packages in time to completion for all simulations.

• Ludwig Geistlinger, Gergely Csaba, Mara Santarelli, Lucas Schiffer, Marcel Ramos, Ralf Zimmer, and Levi Waldron. Towards a gold standard for benchmarking gene set enrichment analysis. Although gene set enrichment analysis has become an integral part of high-throughput gene expression data analysis, the assessment of enrichment methods remains rudimentary and ad hoc. In the absence of suitable gold standards, the evaluation is commonly restricted to selected data sets and biological reasoning on the relevance of resulting enriched gene sets. However, this is typically incomplete and biased towards the individual investigation goals. In this article, we present a curated compendium of 50 expression data sets investigating 34 different human diseases. The compendium features microarray and RNA-seq measurements, and each data set is associated with a precompiled GO/KEGG relevance ranking for the corresponding disease under investigation. We perform a comprehensive assessment of 20 major enrichment methods based on the benchmark set, thereby identifying methods that accurately recover the a priori defined relevance rankings. The compendium is embedded in a directly executable benchmark system, the R / Bioconductor GSEABenchmarkeR package, allowing straightforward execution on additional enrichment methods.

• Nima Hejazi*, Alan Hubbard, Mark van der Laan. Data-Adaptive Estimation and Inference for Differential Methylation Analysis. DNA methylation is amongst the best studied of epigenetic mechanisms impacting gene expression. While much attention has been paid to the proper normalization of bioinformatical data produced by DNA methylation assays, linear models remain the current standard for analyzing post-processed methylation data, for the ease they afford for both statistical inference and scientific interpretation. We present a new, general statistical algorithm for the model-free estimation of the differential methylation of DNA CpG sites, complete with straightforward and interpretable statistical inference for such estimates. The new approach leverages variable importance measures, a class of parameters arising in causal inference, in a manner that facilitates their use in obtaining targeted estimates of the importance of each CpG site. The proposed procedure is computationally efficient and self-contained, incorporating techniques to isolate a subset of candidate CpG sites based on cursory evidence of differential methylation and providing a multiple testing correction that appropriately controls the False Discovery Rate in such multi-stage analysis settings. The effectiveness of the new methodology is demonstrated by way of data analysis with real DNA methylation data, and a recently developed R package (methyvim; available via Bioconductor) that provides support for data analysis with this methodology is introduced.

• Albert Y Zhang, Shian Su, Matthew E Ritchie, and Charity W Law*. Unpacking signal from RNA-seq intron reads using Rsubread and limma packages. RNA-seq datasets contain up to millions of intron reads per library. These reads are typically removed from downstream analysis without even considering the proportion at which they contribute to total reads. By default only reads overlapping annotated exons are thought to be informative since mature mRNA is assumed to be the major component sequenced, especially when examining poly(A) RNA samples. Using Bioconductor packages, Rsubread and limma, we show that intron reads contain signal that is biologically relevant. Multi-dimensional scaling plots show that samples separate into biological and experimental groups using conservative intron counts, where the degree of separation is similar to that of exon counts despite there being far fewer intron reads in comparison to exon reads. The coverage of exon and intron regions are assessed for thousands of genes, showing a tendency for an increase in read coverage from 3’ to 5’ in poly(A) RNA samples. Coverage in Total RNA samples tend to be more uniform in appearance. We show that intron signal is prevalent across multiple datasets and discuss the possibility of its origin from pre-mRNA and intron retention. Results presented here can be used in the development of future Bioconductor packages to interrogate different biological aspects relating to intron signal, and inspire modifications to existing methods for RNA-seq analysis.

• Rachael V Phillips*, Alan Hubbard. Data Adaptive Evaluation of Preprocessing Methods using Ensemble Machine Learning. For many types of biological data generated by high-throughput technologies, there is no single gold-standard for converting the raw data into a form that can be analyzed for relationships of the relevant biomarkers to exposures and disease. For example, much of the variation in the raw data generated by Illumina HumanMethylationEPIC and 450K arrays is due to the technicalities of the experimental design (comprising two different assay methods, two different color channels, and batch effects) and potentially less so due to the biological factor(s) of interest. Accordingly, several preprocessing methods have been developed, however it is unclear which combination should be retained in downstream analysis. To address this issue, we have developed a data adaptive methodology that incorporates ensemble machine learning to assess which preprocessing streams generate better signal-to-noise ratio, according to the prediction of positive and negative control variables. We employ this method to select normalizations for EPIC and 450K arrays in a principled way. The results suggest 1) differences in the relative performance of the possible preprocessing choices and 2) that such machine learning approaches can be practically applied to complex omics data to choose among the growing number of choices for preprocessing.

• Steinbaugh MJ*, Kirchner RD, Ho Sui S. bcbioSingleCell: R package for bcbio single-cell analysis. Single-cell RNA sequencing (scRNA-seq) has ushered in a new era of genomics research, enabling researchers to visualize dynamic changes in cell populations, and quantify changes in gene expression at an unprecedented new level of resolution. While this new technology is extremely exciting and empowering, it remains overly challenging for the vast majority of biologists to import and analyze their results with confidence. The bcbio toolkit addresses this problem, offering native best-practice support for quantification of multiple barcoded droplet platforms, including inDrops, Drop-seq, 10X Genomics Chromium, and Illumina SureCell. The corresponding R package, bcbioSingleCell, provides support for easy loading of results into the SingleCellExperiment container class. Using the standardized SingleCellExperiment container enables researchers to efficiently pass their data to other single-cell packages available on Bioconductor. Additionally, the package offers a suite of quality control functions optimized for low quality cellular barcode removal, along with clustering and visualization functions that integrate with the Seurat toolkit. Differential expression analysis is supported and utilizes the zero-inflated negative binomial model provided by zinbwave, thus unlocking downstream testing with the robust DESeq2 or edgeR RNA-seq packages.

• Righelli D*, Koberstein J, Gomes B, Zhang N, Angelini C, Peixoto L, Risso D. Differential Enriched Scan 2 (DEScan2): a fast pipeline for broad peak analysis. We present DEScan2 a novel bioconductor package for the analysis of Sono-Seq/Atac-Seq data, with the aim to facilitate the investigation of broad peak regions data.

  The method consists of three main steps: 1) a peak caller, 2) a peak filtering and 3) a method to efficiently compute a count matrix of the filtered peaks.

  The peak caller in step 1) is a standard moving window scan that compares the counts within a sliding window to the counts in a larger region outside the window, using a simple Poisson likelihood, providing a final z-score for each peak. However, the package can work with any external peak caller returning results in terms of bed files, indeed the package provides additional functionalities to load bed files of peaks and handle them as GenomicRanges structures.

  The filtering step 2 is aimed to determine if a peak is a “true peak” on the basis of its replicability in other samples. Basing on this idea, we developed the filtering step to filter out those peaks not present in at least a user given number of samples. A further threshold can be used over the peak score.

  Finally, the third step produces a count matrix where each column is a sample and each row a filtered peak computed in the filtering step. The value of the matrix cell is the number of reads for the peak in the sample.

  Furthermore, our package provides several functionalities for GenomicRanges data structure handling. One over the others gives the possibility to split a GenomicRanges over the chromosomes to speed-up the computations parallelizing them over the chromosomes.

• Love MI*, Hickey P, Soneson, C, and Patro R. Automatic metadata propagation for RNA-seq. tximeta performs numerous annotation and metadata gathering tasks on behalf of users during the import of transcript quantifications from Salmon into R / Bioconductor. The key idea within tximeta is to store a signature of the transcriptome sequence, computed and stored by the index and quant functions of Salmon. This signature acts as the identifying information for later building out rich annotations and metadata in the background, on behalf of the user. This should greatly facilitate genomic workflows, where the user can immediately begin overlapping their transcriptomic data with other genomic datasets, e.g. epigenetic tracks such as ChIP or methylation, as the data has been embedded within an organism and genome context, including the proper genome version. We seek to reduce wasted time of bioinformatic analysts, prevent costly bioinformatic mistakes, and promote computational reproducibility by avoiding situations of annotation and metadata ambiguity, when files are shared publicly or among collaborators but critical details go missing.

• Adithya M, Bhargava A, Wright E*. Improving the accuracy of taxonomic classification for identifying taxa in microbiome samples. It has become increasingly clear that the microbiome is an essential component of human and ecosystem health. Microbiome studies frequently involve sequencing a taxonomic marker, such as the 16S rRNA or ITS, to identify the microorganisms that are present in a sample of interest. Here I will describe a new method, named IDTAXA, for taxonomic classification of marker gene sequences that exhibits a substantially lower error rate than previous approaches. In particular, IDTAXA avoids misclassifying sequences belonging to novel taxonomic groups that are not represented in existing taxonomic databases, which is the predominant type of error made by current classifiers. For example, the popular RDP Classifier incorrectly assigns 26.0% of novel 16S rRNA sequences to an existing taxonomic group when the organism actually belongs to a novel taxonomic group. In contrast, IDTAXA only incorrectly classifies 13.6% of such sequences, while correspondingly improving on the fraction of sequences correctly classified to known taxonomic groups. This has a major impact on the interpretation of microbiome data because many microbial communities contain a large fraction of previously undescribed microorganisms that are not yet represented in taxonomic databases. Furthermore, we find that many taxonomic databases contain a considerable number of misclassified sequences that can corrupt the classification process. IDTAXA is able to automatically identify errors in the training taxonomy so that users are able to take corrective action. This enables us to systematically contrast existing taxonomic databases and make recommendations for their use. Collectively, these improvements often lead to substantially different classifications on real microbiome data, which may considerably alter its interpretation. IDTAXA is available as part of the DECIPHER package in R (http://DECIPHER.codes).

• Innes BT* and Bader GD. scClustViz - Single-cell RNAseq Cluster Assessment and Interactive Visualisation. Single-cell RNA sequencing is becoming an increasingly popular technology, used both in large multi-centre projects, and for more directed biological experiments. A common purpose for applying this technology is the in silico classification of cell types in a tissue. This is generally done using one of a myriad of clustering algorithms, based on the assumption that cells within a cell type are share similar transcriptomes, which are distinct from other cell types in the tissue. However, nearly all clustering algorithms have tunable parameters which affect, either directly or indirectly, the number of clusters they will return from the data.

  The R Shiny software tool outlined here provides a simple interactive interface for assessing the biological relevance of clustering results. Given that cell types are expected to have distinct gene expression patterns, it uses differential gene expression between clusters as a metric for assessing overfitting of clustering (Yuzwa et al., Cell Reports 2017. DOI:10.1016/j.celrep.2017.12.017). Along with this, it also provides interactive visualisation of: cluster-specific distributions of technical factors and other metadata; cluster-wise gene expression statistics to simplify annotation of cell types and identification of marker genes; and gene expression distributions over all cells.

  Interactive user interfaces for single-cell RNAseq analysis already exist, but this tool fills a distinct niche, as it is explicitly designed to assist in the biological interpretation of clustering solutions. Since it is meant to be included in analysis workflows, it has intentionally been built to be easily customized by the bioinformatician, as well as used by the non-technical biologist.

  This tool provides an interactive interface for visualisation, assessment, and biological interpretation of cell-type classifications in single-cell RNAseq experiments that can be easily added to existing analysis pipelines, allowing non-technical biologists easier access to their data.