Menu

Newsletters

2016

2015

2014

 

Bioconductor Newsletter

posted by Valerie Obenchain, April 2015

On April 17, the release of Bioconductor 3.1 will mark the 24th release of the software. The project started in 2001 with the first svn commits made in May of that year:

r3      rgentlem      2001-05-25 14:08:57 -0700 (Fri, 25 May 2001)
r2      rgentlem      2001-05-25 08:28:31 -0700 (Fri, 25 May 2001)
r1      (no author)      2001-05-25 08:28:31 -0700 (Fri, 25 May 2001)

At the time of the first official Bioconductor manuscript in 2004 the project consisted of

"... more than 80 software packages, hundreds of metadata
packages and a number of experimental data packages ..."

Eleven years later (and after more than 100000 svn commits) Bioconductor hosts over 990 software, 900 annotation and 230 experimental data packages.

Another quote from the 2004 paper shows that, fortunately, not everything has changed,

"... The group dynamic has also been an important factor in the success of 
Bioconductor. A willingness to work together, to see that cooperation and 
coordination in software development yields substantial benefits for the 
developers and the users and encouraging others to join and contribute to 
the project are also major factors in our success. ..."

This issue looks at the growing role of proteomics in Bioconductor and the use of web sockets to bridge the gap between workspace data and interactive visualization. We re-visit Docker with use cases in package development and managing system administration tasks. We also have a section on new and notable functions recently added to base R and Bioconductor.

Contents

Proteomics in Bioconductor

The diversity of proteomics analysis available in Bioconductor continues to grow steadily and the devel branch now hosts 68 proteomic-based software packages. Many individuals have contributed to this area in the form of packages, web-based workflows and course offerings. One very active member is Laurent Gatto, head of the Computational Proteomics Unit at the Cambridge Centre for Proteomics. On a day to day basis he is responsible for developing robust proteomics technologies applicable to a wide variety of biological questions.

Laurent is the author of many Bioconductor packages, including the new ProtGenerics. Similar in concept to the more general BiocGenerics package, ProtGenerics provides a central location where proteomic-specific S4 generics can be defined and reused. He has produced course materials and tools to help newcomers get started including the detailed proteomics workflow on the web site and a two publications titled Using R and Bioconductor for proteomics analysis and Visualisation of proteomics data using R and Bioconductor. These publications have a companion experimental data package, RforProteomics which illustrates data input/output, data processing, quality control, visualisation and quantitative proteomics analysis within the Bioconductor framework. Since the first release, RforProteomics has benefited from contributions from additional developers.

The RforProteomics data package has 4 vignettes:

  • The first vignette offers a current perspective on proteomics in Bioconductor and is in poster format. It gives an overview of the Bioconductor proteomics infrastructure and mass spectrometry analysis. Topics covered include raw data manipulation, identification, quantitation, MS data processing, visualization, statistics and machine learning.

  • Also in poster format, the RforProteomics BioC2013 vignette is specific to the Using R and Bioconductor for proteomics analysis publication. Special attention is given to labelled vs label-free quantitation and Bioconductor packages that offer these methods.

  • Using R and Bioconductor for Proteomics Data Analysis includes code executed in the Using R and Bioconductor for proteomics analysis publication.

  • Visualisation of proteomics data using R and Bioconductor includes the code from the Visualisation of proteomics data using R and Bioconductor publication.

Many proteomic packages are worthy of mention in the Bioconductor repository. Here we highlight a few that have played a primary role in the growing infrastructure.

Packages mzR and mzID read and parse raw and identification MS data. The former is an R interface to the popular C++ proteowizard toolkit.

Identification methods are offered in rTANDEM and MSGFplus (and its shiny interface MSGFgui) and quantitation methods can be found in MSnbase and isobar (isobaric tagging and spectral counting methods), synapter, xcms and MALDIquant packages (label-free). Statistical modelling and machine learning are offered in MSnbase, isobar, MSstats and msmsTests.

The rpx package provides an interface to the ProteomeXchange infrastructure, which coordinates multiple data repositories of MS-based proteomics data.

The pRoloc package contains methods for spatial proteomics analysis, e.g., machine learning and classification methods for assigning a protein to an organelle.

The relatively new Pbase package contains a Proteins class for storing and manipulating protein sequences and ranges of interest. The package has multiple vignettes on coordinate mapping. One addresses mapping proteins between different genome builds and the other mapping from protein to genomic coordinates.

Web Sockets

Overview

As our ability to generate volumes of sequencing data grows so does the need for effective visualization tools. Tools that can summarize large quantities of information into digestible bits and quickly identify unique features or outliers are important steps in any analysis pipeline. In the R world we have seen an increase in the use of web sockets to provide an interactive link between data in the workspace and exploration in the browser.

The analysis capabilities of R make it a good fit for the rich and interactive graphics of HTML5 web browsers. The WebSocket protocol enables more interaction between a browser and a web site, facilitating live content and the creation of real-time graphics.

Web sockets are often described as “a standard for bi-directional, full duplex communication between servers and clients over a single TCP connection”. These characteristics offer several advantages over HTTP:

  • ‘bi-directional’ means either client or server can send a message to the other party. HTTP is uni-directional and the request is always initiated by the client.

  • ‘full duplex’ allows client and server to talk independently. In the case of HTTP, at any given time, either the client is talking or the server is talking.

  • Web sockets open a ‘single TCP connection’ over which the client and server communicate for the lifecycle of the web socket connection. In contrast, HTTP typically opens a new TCP connection for each round trip; a connection is initiated for a request and terminated after the response is received. A new TCP connection must be established for each request/response. The opening and closing creates overhead, especially in the case where rapid responses or real time interactions are needed.

For those interested, this blog post provides more in-depth details and benchmarking against REST.

Applications in Shiny and epivizr

The Shiny package created by the RStudio team pioneered the use of web sockets in R. Shiny enables the building of interactive web applications from within an R session. Popular applications are interactive plots and maps that allow real-time manipulation through widgets. The workhorse behind Shiny is the httpuv package, also authored by RStudio. httpuv provides low-level socket and protocol support for handling HTTP and WebSocket requests within R.

The httpuv infrastructure is also used by the epivizr package. In this application, web sockets create a two-way communication between the R environment and the Epiviz visualization tool. Objects available in an R session can be displayed as tracks or plots on Epiviz.

BrowserViz

A slightly different approach is taken in the new BrowserViz package by Paul Shannon. This application provides access to both the browser and an active R prompt.

The BrowserViz package contains the BrowserViz class whose main purpose is to provide the necessary R, Javascript websocket and JSON infrastructure for communication. By loosely coupling R and the browser the two environments are linked but kept maximally ignorant of each other; only simple JSON messages pass back and forth with no HTML, CSS or Javascript. The result is access to the interactive graphics of a web browser in conjunction with an active R session.

The companion Javascript library, BrowserViz.js, is also included in the package. The combination of library and base class provides the infrastructure necessary for any BrowserViz-style application. The BrowserVizDemo and RCyjs packages build on the BrowserViz class and will be available in the Bioconductor 3.2 release. BrowserVizDemo is a minimal example of interactive plotting and selection of xy points using the popular d3.js library. The more full featured RCyjs provides interactive access to the full power of Cytoscape.js, a richly featured browser-based network visualization library.

More details on the BrowserViz class and applications can be found in the package vignette.

Infrastructure

Changes in AnnotationHub

This quarter Marc and Sonali continued their work on AnnotationHub. Several new resources were added and the display method and search navigation were substantially reworked.

New resources:

  • TF-Target Gene Files from the PAZAR public database (available as GRanges objects)
  • BioPAX files (Level1 and Level2) from the NCI Pathway Interaction Database (available as biopax objects)
  • Background file for ChEA required for the command line version of ChEA (available as data.frame object)
  • GTF files from Ensembl release 76 to 79 (available as GRanges object)
  • Expression Set of raw read counts for the GSE62944 dataset from GEO: (available as ExpressionSet object)

An improved show method and more flexible data retrieval make interacting with the 18900+ files straightforward. Sonali has a new AnnotationHub video where she gives a tour of the resource with tips and tricks for data access.

Code below was generated with AnnotationHub version 1.99.75. The show method now list fields common for subsetting up front, e.g., providers, species and class of R object.

> library(AnnotationHub)
> hub <- AnnotationHub()
> hub
AnnotationHub with 18992 records
# snapshotDate(): 2015-03-12 
# $dataprovider: UCSC, Ensembl, BroadInstitute, NCBI, Haemcode, dbSNP, Inpar...
# $species: Homo sapiens, Mus musculus, Bos taurus, Pan troglodytes, Danio r...
# $rdataclass: GRanges, FaFile, OrgDb, ChainFile, CollapsedVCF, Inparanoid8D...
# additional mcols(): taxonomyid, genome, description, tags, sourceurl,
#   sourcetype 
# retrieve records with, e.g., 'object[["AH169"]]' 

            title                                         
  AH169   | Meleagris_gallopavo.UMD2.69.cdna.all.fa       
  AH170   | Meleagris_gallopavo.UMD2.69.dna.toplevel.fa   
  AH171   | Meleagris_gallopavo.UMD2.69.dna_rm.toplevel.fa
  AH172   | Meleagris_gallopavo.UMD2.69.dna_sm.toplevel.fa
  AH173   | Meleagris_gallopavo.UMD2.69.ncrna.fa          
  ...       ...                                           
  AH28575 | A500002_Erg.csv                               
  AH28576 | A500005_Erg.csv                               
  AH28577 | A500001_IgG.csv                               
  AH28578 | A500004_IgG.csv                               
  AH28579 | GSM730632_Runx1.csv

Tab completion on a hub object lists all fields available for subsetting:

> hub$
hub$ah_id         hub$dataprovider  hub$taxonomyid    
hub$description   hub$rdataclass    hub$sourcetype    
hub$title         hub$species       hub$genome        
hub$tags          hub$sourceurl

Quick discovery of file type and provider:

> sort(table(hub$sourcetype), decreasing=TRUE)

          BED         FASTA    UCSC track           GTF NCBI/blast2GO 
         7855          3876          2208          1606          1145 
        Chain           CSV           VCF        BigWig    Inparanoid 
         1113           406           316           315           268 
       TwoBit  BioPaxLevel2         RData        BioPax         GRASP 
          144             6             4             3             1 
       tar.gz           Zip 
            1             1 

> sort(table(hub$dataprovider), decreasing=TRUE)

                            UCSC                          Ensembl 
                            8746                             4590 
                  BroadInstitute                             NCBI 
                            3146                             1145 
                        Haemcode                            dbSNP 
                             945                              316 
                     Inparanoid8                            Pazar 
                             268                               91 
NIH Pathway Interaction Database                        EncodeDCC 
                               9                                5 
                          RefNet                             ChEA 
                               4                                1 
                             GEO                            NHLBI 
                               1                                1

Given the volume and diversity of data available in the hub we encourage using these files as sample data before creating your own experimental data package.

For example, to get an idea of available GRCh37 FASTA from Ensembl:

>  hub[hub$sourcetype=="FASTA" & hub$dataprovider=="Ensembl" & hub$genome=="GRCh37"] 
AnnotationHub with 42 records
# snapshotDate(): 2015-03-26 
# $dataprovider: Ensembl
# $species: Homo sapiens
# $rdataclass: FaFile
# additional mcols(): taxonomyid, genome, description, tags, sourceurl,
#   sourcetype 
# retrieve records with, e.g., 'object[["AH18924"]]' 

            title                                    
  AH18924 | Homo_sapiens.GRCh37.73.cdna.all.fa       
  AH18925 | Homo_sapiens.GRCh37.73.dna_rm.toplevel.fa
  AH18926 | Homo_sapiens.GRCh37.73.dna_sm.toplevel.fa
  AH18927 | Homo_sapiens.GRCh37.73.dna.toplevel.fa   
  AH18928 | Homo_sapiens.GRCh37.73.ncrna.fa          
  ...       ...                                      
  AH21181 | Homo_sapiens.GRCh37.72.dna_rm.toplevel.fa
  AH21182 | Homo_sapiens.GRCh37.72.dna_sm.toplevel.fa
  AH21183 | Homo_sapiens.GRCh37.72.dna.toplevel.fa   
  AH21184 | Homo_sapiens.GRCh37.72.ncrna.fa          
  AH21185 | Homo_sapiens.GRCh37.72.pep.all.fa

Advanced developers may be interested in writing a ‘recipe’ to add additional online resources to AnnotationHub. The process involves writing functions to first parse file metadata and then create R objects or files from these metadata. Detailed HOWTO steps are in the AnnotationHubRecipes vignette.

Rhtslib package

Nate recently completed work on the Rhtslib package which wraps the htslib C library from Samtools. The plan is for Rhtslib to replace the Samtools code inside Rsamtools. Rhtslib contains a clean branch of htslib directly from Samtools, including all unit tests. This approach simplifies maintenance when new versions or bug fixes become available. The clean API also promises to make outsourcing to the package more straightforward for both Rsamtools and others wanting access to the native routines.

htslib was developed with a ‘linux-centric’ approach and getting the library to build across platforms (specifically Windows) was a challenge. To address this, Nate chose to use Gnulib, the GNU portability library. Briefly, Gnulib is a collection of modules that package portability code to enable POSIX-compliance in a transparent manner; the goal being to supply common infrastructure to enable GNU software to run on a variety of operating systems. Modules are incorporated into a project at the source level rather than as a library that is built, installed and linked against.

Incorporating Gnulib involves (at minimum) the following steps:

  • adapt the project to use Autoconf and Automake
  • identify and import relevant Gnulib modules using gnulib-tool
  • add #include "config.h" to source files
  • remove (now unnecessary!) preprocessor complier/platform tests from source

For more on specific functions available in Rhtslib see the Samtools docs or the API-type headers in the package, faidx.h, hfile.h, hts.h, sam.h, tbx.h and vcf.h. Headers are located in Rhtslib/src/htslib/htslib or if the package is installed,

library(Rhtslib)
system.file(package="Rhtslib", "include")

Reproducible Research

The course materials web page has links to several resources including slides, presentations and packages. Recently Dan started adding an “AMI” link for courses that use them. The AMI contains the packages, sample data and exact version of R/Bioconductor used. This is a convenient, portable way to ensure reproduciblity. One can imagine using an AMI or Docker container to capture the state of a research project or publication which can be easily shared with colleagues.

Developing with Docker

Elena Grassi is a Ph.D. student in Biomedical Sciences and Oncology in the Department of Genetics, Biology and Biochemistry at the University of Torino. Her research focuses on transcriptional and post transcriptional regulation with special interest in transcription factors and the alternative polyadenylation phenomenon.

With a background in computer science she is involved in developing computational pipelines and tools and is the author of Bioconductor packages roar (preferential usage of APA sites) and MatrixRider (propensity of binding protein to interact with a sequence). Elena was one of the first to try out the Docker containers and found them useful for both package development and system administration tasks. I asked a few questions about her experience and got some interesting answers.

What motivated you to try Docker when developing MatrixRider?

I heard about docker from some friends last year and I was eager to try it. During the New Year’s Eve holidays I decided to start using it with R / Bioconductor to run different versions on our computational server without adding burden to the sysadmin work. I started with mere curiosity fiddling with rocker and, eased by the holiday laziness, I stopped with the idea to begin working on some ad hoc R \ Bioconductor containers in January. Imagine my happiness when I read in the newsletter about the brand new Bioconductor docker containers: they were ready for me :). I decided to use them to develop MatrixRider as long as I needed to have working versions both for release and devel. I work on different computers and using the Bioconductor devel_sequencing container freed me completely from the procedure of getting the source, building, and installing all needed packages. Besides this advantage using docker made me sure that the package I was developing did not have any dependencies on my local system libraries that would not be available in a clean installation. This was my first package containing C code and it was nice to be sure.

Which image did you use, base, core, sequencing, … ?

Mainly devel_sequencing to start with a fully-fledged working environment. I had to install some other packages (TFBSTools and JASPAR2014) and it worked flawlessly with biocLite().

Any unanticipated pros/cons of developing in these containers?

No. I think that I will continue using the devel containers to develop and maintain packages.

Describe how Docker was useful for managing multiple R versions on your computational server. Was this for multiple users or just yourself?

Right now I’m the only one that needs devel so the version management was for myself. Eventually I would like to set it up on our server and have it working for multiple users but it will take a little work to integrate it with our “pipeline management system”.

In the past we have had up to three different R versions, one from the package management system of our distribution, Debian, and two compiled ad hoc. Teaching new students how to reach them and the related library paths has been hard - I am pretty sure docker will give a huge hand in these situations, helping also in tracking which versions of packages were used to perform certain analyses.

New and Noteworthy

A number of functions added to R (3.2) and Bioconductor (3.1) this quarter have potential for wide-spread use. I thought they were worth a mention.

  • base::lengths()

    Computes the element lengths of a list object. In Bioconductor, S4Vectors::elementLengths performs the same operation on List objects. (contributed by Michael Lawrence)

  • base::trimws()

    Removes leading or trailing whitespace from character strings. (contributed by Kurt Hornik)

  • utils::methods()

    This function previously worked on S3 generics only and has been enhanced to also handle S4. (enhanced by Martin Morgan)

    > library(Rsamtools)
> methods("scanBam")
[1] scanBam,BamFile-method    scanBam,BamSampler-method
[3] scanBam,BamViews-method   scanBam,character-method 
see '?methods' for accessing help and source code
Warning message:
In findGeneric(generic.function, envir) :
  'scanBam' is a formal generic function; S3 methods will not 
   likely be found

> methods(class = "BamFile")
 [1] $                   $<-                 asMates            
 [4] asMates<-           close               coerce             
 [7] countBam            filterBam           indexBam           
[10] initialize          isIncomplete        isOpen             
[13] obeyQname           obeyQname<-         open               
[16] path                pileup              qnamePrefixEnd     
[19] qnamePrefixEnd<-    qnameSuffixStart    qnameSuffixStart<- 
[22] quickBamFlagSummary scanBam             scanBamHeader      
[25] seqinfo             show                sortBam            
[28] testPairedEndBam    updateObject        yieldSize          
[31] yieldSize<-        
see '?methods' for accessing help and source code

  • GenomicFeatures::transcriptLengths()

    Computes transcripts lengths in a TxDb object with the option to include / excluded coding and UTR regions. (contributed by Hervé Pagès)

  • BiocParallel::bpvalidate()

    Flags undefined symbols in functions intended for parallel, distributed memory computations. (contributed by Martin Morgan, Valerie Obenchain)

  • BiocInstaller::biocLite()

    Now capable of installing git repositories. When the ‘pkg’ argument contains a forward slash, e.g., “myRepo/myPkg”, it is assumed to be a repository and is installed with devtools::install_github. (contributed and enhanced by Martin Morgan)

Project Statistics

Website traffic

The following compares the number of sessions and new users from the first quarter of 2015 (January 1 - March 30) with the first quarter of 2014. Sessions are broken down by new and returning visitors. New visitors correspond to the total new users.

First Quarter Website Traffic 2015 vs 2014
Sessions: Total 24.03% (339,283 vs 273,559)
Sessions: Returning Visitor 21.42% (213,848 vs 176,128)
Sessions: New Visitor 28.74% (125,435 vs 97,431)
New Users 28.74% (125,435 vs 97,431)


Statistics generated with Google Analytics.

Package downloads and new submissions

The number of unique IP downloads of software packages for January, February and March of 2015 were 31720, 31956, and 38379, respectively. For the same time period in 2014, numbers were 29690, 28993 and 34634. Numbers must be compared by month (vs sum) because some IPs are the same between months. See the web site for a full summary of download stats.

A total of 55 software packages were added in the first quarter of 2015 bringing counts to 991 in devel (Bioconductor 3.2) and 936 in release (Bioconductor 3.1).

Upcoming Events

See the events page for a listing of all courses and conferences.

Acknowledgements

Thanks to Laurent Gatto, Elena Grassi and Paul Shannon for contributing to the Proteomics, Docker and Web Sockets sections. Also thanks to the Bioconductor team in Seattle for project updates and editorial review.

Send comments or questions to Valerie at [email protected].