I'm at the 2017 Bioinformatics Open Source Conference (BOSC) in Prague. These are my notes from the day 1 afternoon sessions on developer tools, reproducibility, data science and visualization. It also includes notes from the keynote by Madeleine Ball on open human genome data.
BOSC is a two day annual conference organized by the Open Bioinformatics Foundation. It's an open community interested in building reusable and interoperable tools for doing biological research. All video from these presentations will be available from the schedule page and slides from the f1000 channel.
Phil Ewels
Phil works at SciLifeLab and is part of National Genomics Infrastructure in Sweden. Movitated by moving from command line based QC done on their many samples. Developed MultiQC, which collects multiple samples and multiple QC methods into a single report. It provides an at a glance overview of many metrics and is a brilliant way to present QC. bcbio integrates MultiQC and it's been a huge improvement. It can scale to thousands of samples by having static plots instead of interactive; large tables turn into plots. On the backend MultiQC has great configuration and it is easy to plug in new tools. At the Codefest, lots of new users got involved and contributed significant improvements to MultiQC in two days. Wow, there is a plugin for MultiQC to interact with ClarifyLIMS: pulls in sample metadata and project information from the LIMS using the API. Coming soon, a new project called MegaQC that collects results from many MultiQC runs and visualize trends, uses a local server with database and website.
Alexander S Rose
NGL is a WebGL based viewer for molecular graphics. Easy to inject directly into web pages, can use javascript to tweak and do calculations on the client side. Handles a wide variety of molecular formats and tons of representations.
Aditya Bharadwaj
GRAPHSPACE is a collaborative way to share network results across multiple groups. Motivation: due to the complexity it is currently quite hard to do. GRAPHSPACE allows sharing with collaborators or openly, 300+ users sharing already so really filling a niche in the biological community. Nice integration with Cytoscape web. Really solid project wishing I knew more about networks.
flowcells Tim Booth
Describes how Illumina patterned flowcells work, in terms of technical problems that can happen during sequencing. Well hopping involves sequences leaking into surrounding cells. This creates technical duplicates which over-represents the amount of sequence for RNA-seq or variant calling. Standard approach is Picard mark duplicates after alignment. Their solution is a quicker and works directly from bcl files. Code available on GitHub
Monther Alhamdoosh
Gene set testing: run an experiment, get a bunch of outputs, then try to organize into groups that provide biological meaning. Different methods end up with different gene sets -- hard to identify what is correct, danger of trying multiple methods until you get something that "makes sense." Ran 12 different methods and then compared them. EGSEA (pronounce eg-zee) supports 12 methods and runs some limma magic to combine into a final gene set combining the information from all of them. It's a R package available from bioconductor.
Timo Sachsenberg
Main high throughput method in proteomics/metabolomics in LC-MS, which separates and quantifies analytes. OpenMS is an open source framework to compute on mass spec output. Everything open source and well tested, been developed for 10+ years. Great example of long running successful project. Has a workflow, tool and library layer to access -- run something pre-built, combine tools however you want, or code against the API directly. 185 tools that cover almost everything you'd want to do with mass spec.
Brad Chapman
I talked about how the Common Workflow Language enables bcbio to run on multiple platforms.
Kees van Bochove
cBioPortal is a cancer genomics tool, currently supported and developed by The Hyve. GSVA is a method to turn genes into gene sets for analysis. This is fully integrated into cBioPortal, supporting queries and heatmaps. cBioPortal has full sets of open public datasets like TCGA data.
Björn Grüning
How can we make a FAIR (Find, Accessible, Interoperable, Reusable) set of tool dependencies. Workflow -- development of code, packaging, then deploying. There was an explosion of package managers: language specific (R, Python, Perl..) and system specific (deb, RPM) plus higher level containers (Docker). The current biology community solution to avoid this: conda a cross platform, language agnostic installation method. Demonstrates how to build a tool with it. Bioconda is a community building scientific tools. It has an automated, unified build environment and integrates with other conda communities like conda-forge. Huge incredible community maintaining tools. Also autogenerates containers from conda recipes, making Docker and Singularity containers available. Everything happens automatically via TravisCI. Works with Biocontainers which provides standards for running Docker tools.
Ricardo Wurmus and Pjotr Prins
Ricardo talks about the difference between sysadmins and users. Users have ad-hoc, volatile and provides no backups/updates. If you need 100% reproducibility it's hard to isolate from the system even with bioconda, easybuild. They're not properly abstracted. Docker/containers provide some solution but hard to compose. GNU Guix is a functional packaging manager which always provides 100% reproducibility. Level of abstraction matters, reproducible and safe experimentation, Guix makes sharing easy, ways to use Guix without root. Guix installs conda.
John Chilton
John follows up on the bioconda discussion with how it enables computational platforms. HPC-ready, cross-platform, easy to manage/maintain multiple versions of the same tool. Galaxy provides the ability to have explicit containers using Biocontainers. Enables cool things like running Kubernetes and Singularity. cwltool now has configurable dependency resolution. Shows example of an amazing workflow porting: an 11 tool ChIP-seq workflow that got ported from AWS + Docker to HPC without Docker.
Revitalizing a classic bioinformatics tool using modern technologies: the case of the Cytoscape Project
Keiichiro Ono
Cytoscape is a 15 year old project, implemented in Java. Today, v3.5 released and community sill growing, 300+ associated apps. It's not an ecosystem of tools. But web is the new platform for platform, how can it move there without getting rid of the existing ecosystem. Now has a REST API and CyComponents, which are react-based reusable UI/dataviz components. Pilot project building off this: CyIndex 2.0 -- a repository for biological networks.
Olga Vrousgou
SPOT is the Sample, Phenotype and Ontologies team at EBI trying to help you annotate your data with ontologies. Ontologies help with data integration. ZOOMA provides automated mapping to onologies. When it's wrong, OxO cross reference ontologies. The OLS (ontology lookup service) lets you find terms when cross referencing doesn't help. Webulous lets you create ontology terms when non exist. All tools on GitHub.
Christian Brueffer
Biopython project is an original Open Bioinformatics foundation project from 1999. Walk through all of the new releases and work this year. Moved to the BSD license, which is helpful since the Biopython license was custom and essentially BSD. Three new releases this year with lots of new contributors. Working on a modernized build process with python wheel format. Lots of use of continuous integration: TravisCI, Appveyor for windows testing,
Madeleine Ball
Madeleine starts by describing her background: fixed DNA representation on Wikipedia, one laptop per child project, more Wikipedia open work sharing knowledge. At the same time, working in George Church's lab. She got involved with the Personal Genome Project -- open sequence data plus open and extensive phenotype data. Director of Research for several years, worked a lot on ethical and legal issues. Had full engagements with participants including returning personal genome. Most genomes we share are not available to participants. Why don't we do this as scientists? PGP had amazing communication with participants: community forum, conference plus discussion directly with Madeleine. Contrast with traditional research projects.
Scary date: how worried should you be about sharing your genome data? Some possibilities: health issues, ancestry, re-identification with Y chromosomes. Not a lot of easy answers but lessons to learn. Genomes taught more general lessons. Uses example of location data of her life in NYC -- this also re-identifies you and could be potentially problematic. Not genome specific.
PGP consent -- very difficult to do correctly. Had an extensive consent process with a quiz, which makes people actually understand what they are consenting too. In the end: people like to share and like to be asked. They have >3,500 individual with public genome and genotyping data. There is always data that goes too far for you: examples of location, Google Search data.
The current research model: original data holders are universal gatekeepers. Can't talk to the original individuals either without okay/consent/payment to the original data holder.
Other approaches: what if everyone was in the same place and could hear you? What is it was easy to return data? What is users could contribute back? Good example of adding a HealthKit app for OpenHumans. Idea: have all the data available with individuals and allow them to share, make it available as they want. This allows you share data across studies over time.
Time for a concrete example. Dana Lewis and OpenAPS project to monitor her Type I diabetic Artifical Pancreas System. NightScout is software that monitors nighttime blood sugar to prevent it from being too low because of 1:20 chance of dying in sleep as Type I diabetic.
OpenHumans offers $5000 grants to do work with Open Humans data and also has a job available for a programmer. It's a great chance to contribute to open, innovative work.