I’m at the 2016 Bioinformatics Open Source Conference (BOSC) in Orlando, Florida. BOSC is a two day community conference devoted to open scientific development communities. Nomi Harris starts the day off with an introduction to the 17th annual BOSC. The theme this year is about connecting Communities of Communities and Nomi emphasizes the importance of bringing together multiple independent groups to create a larger community that can solve important problems. This is the key goal of BOSC and why we come together to share and learn from each other.
Jennifer Gardy – The open-source outbreak: can data prevent the next pandemic?
Jennifer starts the conference off talking about the roll of open data in infectious disease. She has some great stories about presentations from BOSC 2004 and trips to Orlando when she was in high school. On the science side, Jennifer talks about using genomic data to understand where new diseases come from and how we can understand their spread. 56 million deaths a year and 1/3 of these are due to infectious disease; 5x higher death rate in low income countries. Tuberculosis has been found in mummies from 2050BC, so we have both old and new diseases to deal with. Old diseases can change and become resistant, so lots of things to worry about. Beautiful map of where diseases emerge on a global scale. We know where they come from but are not looking for new diseases in a systematic way. Most locations do not have systems for collecting and sharing data, mostly resource constrained. Good example is Ebola: December 6th first deadh, March 22nd first pro-med e-mail alert, August declaration of emergency. Public health is quite bureaucratic and most data is kept private to research groups.
Awesome work on crowd sourcing problem: EpiCollect for your phone, EpiHack for in depth problems. Called Digitial epidemiology. HealthMap provides reports on issues located nearby. They identified Ebola before pro-med. nEmesis monitors tweets for food poisoining updates.
Lots of bioinformatics opportunities in disease detection: rapid ID of pathogenes from metagenomic surveillance data. Example of the great ZIBRA project sequencing Zika in Brazil and releasing data in real time.
Shows the incredible John Snow graphic from Cholera outbreak in London. Identified infected water pump. Still useful technique – identify isolates, use molecular epidemiology to id clustered isolates. Then try to find connections between clusters. However, lots of current limitations: you don’t get order/direction of transmission, size and membership of cluster varies, lots of manual work to identify underlying transmission structure. Genomic epidemiology: use genomic sequence to track how it spreads without needing to manually talk to everyone. Data is simple to use – 21 base pairs vary over whole genome in tuberculosis outbreak and can compare only these. Easy to visualize and see sub-groups with high resolution picture. Next work – want to automatically infer transmission from this structure. Infer a transimission tree from phylogenetic tree: uses Beast to draw tree of outbreak, and identifies locations where there are jumps using this to infer an infection network.
Cool examples of open data and analyses available to anyone: Virological, NextFlu. Open-source crowd-sourced analysis of E coli outbreak paper. So incredible, how can the community do it regularly? Challenge is to bridge the evolutionary-math-bioinformatics gap. Better information visualization and interpretation needed.
Ted Liefeld – GenomeSpace: Open source interoperability platform with crowd-sourced analysis recipes
GenomeSpace is an open source tool to connect bioinformatics programs. Provides a standard way to organize files in the cloud, then ship them off to integrated tools. Has publicly shared files. 20 GenomeSpace tools, including cBio portal, Galaxy, GenePattern, ISATools, Cytospace. Great example of connecting tools together and creating community of communities. To integrate a tool it needs to be able to handle authorization, read files from URLs. Have a recipe resource with step-by-step instructions for doing integrative analyses. Awesome community distributed way to document and share.
Michael Crusoe – This is Why We Can Have Nice Things: Getting to 1.0 of the Common Workflow Language
Michael talks about the release of Common Workflow Language (CWL) v1.0. The motivation is that there are many workflow standards and could we move workflows between them. Standards create a surface for collaboration that promotes innovation. Works on both shared-nothing clusters (cloud), academic clusters with shared filesystems. Michael does a great job of explaining the goals of CWL (practical standard) and the community (large set of members) and lessons learned. He also presents an awesome vision of building great full reproducible workflows with the workshope for sustainable software in science.
Dan Leehr – CWL in Practice: Experiences, challenges, and results from adopting Common Workflow Language
Peter Amstutz – Using the Common Workflow Language (CWL) to run portable workflows with Arvados and Toil
Peter works on the Arvados project and will talk about work running pipelines in CWL in multiple environments. What kind of software can we have if we have the baseline assumption that we can move workflows between systems. Can we run an unmodified workflow using completely different workflow software, cloud providers, storage systems and schedulers. Used bcbio to run in two environments: Toil and Arvados. Toil: running on AWS, S3 storage, Mesos scheduler, converts CWL to Toil workflow graph. Arvados is a managed multi-tenant architecture with web workbench, running on Azure, Arvados Keep storage for files, Crunch + SLURM scheduler. Ran and got the exact same outputs. Great demonstration. To ensure this kind of compatibility, there is continuous validation: a CI server and continuously tests every implementation and provide guidance to users of CWL. If you can trust ability to bring your own workflow you can choose the platform that matches your needs. Portable APIs associated with this: GA4GH Tool Registry API. DockStore implements this API and is a usable implementation available now. GA4GH Workflow submission API for further standardization.
John Chilton – Planemo – A Scientific Workflow SDK
Galaxy philosophy on workflows – the most important user is the bench scientist using the GUI. Galaxy will never require an SDK, but the SDKs are rather for bioinformaticians who prefer this approach over the GUI. Planemo (pronounced Plah-nemo – Nemo like the famous fish) is the way to develop Galaxy tools and focuses on developers. planemo creates a profile for testing workflows, then can re-run without needing setup every time. Galaxy’s workflow format is JSON, hard to read and impossible to write. Swapped over to a Format 2 workflow which is very similar to CWL. CWL-inspired and hopefully real CWL soon. Planemo also provides nice facilities to test workflows. CWL and Galaxy: right now CWL tools work with Galaxy tools. No support for CWL workflows yet but hopeful outcome for BOSC 2017. Planemo can lint CWL tools – useful functionality for standard CWL development. John describes other great work to make tool installation easier: bioconda, docker.
Daniel Blankenberg – Sample Size Does Matter: Scaling Up Analysis in Galaxy with Metagenomics
Dan talking about enabling metagenomic work with Galaxy. Handles a whole bunch of standard metagenomic tools. Dan is incredibly fast having trouble keeping up. Handles normalization, metadata, graphs of differentiation between results, integrated Phinch from Holly Bik (last year’s keynote – awesome). Also handles large-scale multiple sample analysis – 500 samples. 5000+ – still under development.
Fabien Campagne – NextflowWorkbench: Reproducible and Reusable Workflows for Beginners and Experts
Nextflow Workbench is an integrated development enviornment for Nextflow. Nice typing system, auto-completion, error highlighting. It’s a GUI environment that makes developers much more productive. Looks like a great environment, built on top of Nextflow so can work on laptops, clusters, or Google cloud. It’s also built in MPS as a cloud language.