VariantDB_Challenge
Follow me on Twitter (@Variant_Chall) to recieve updates on the project!
This repo is designed to serve as a testing framework for different database schemas, architectures, and query formulations within the context of mining genomics data. For some reason, we in the genomics community (with a few outliers) are stuck in a file-based modus operandi. Most people store variants in either a VCF or gVCF. However there are numerous reasons why this is a bad long-term solution.
-
VCF files need to be filtered in many different ways depending on the study question
- For example, if I wanted to exclude variants that change the protein sequence, I would write a filter program on the command-line and then write the subsetted VCF into yet another file. This is a huge burden on data storage, as well as tracking data provenance.
-
It's just not a scalable strategy
- I have a hard enough time keeping track of where all my files are from this year, let alone last year. Instead many of us have to keep separate files that track where all of our other files are located.
-
VCF files do not store metadata
- VCF files are OK for storing genetic variants, but the lack information about the samples they represent. For example, simple metadata such as which samples are cases or controls, how old was the patient, etc. is not available in the current specification.
It's pretty clear, we need to move to a database if we want to stay on top of the problem of data growth that is to be expected. However, given the scale and nature of genomics data, this isn't a simple solution. Those that aren't in this field may ask,
Why not just throw all your data into a MySQL database? Won't that work?
Technically, yes but pragmatically no. The key limitation is that the RDBMS approach scales up rather than scaling out for this particular use case. Genomics is somewhat heterogeneous, constantly evolving and growing, and needs to leverage distributed computing and storage. Finding a solution that takes advantage of these key concepts is what this project is all about!
The future of genomics depends on our ability to leverage distributed storage and computing, so what database architecture is best? Let's find out by addressing the following challenges:
Challenge #1. The multiple source integration problem.
Challenge #2. The continuous integration problem.
Challenge #3. The cost projection problem.
The purpose of this challenge is not an ideological one, rather an empirical one. We are looking for submissions from experts of various architectures to submit solutions to address these challenges. If you think you have a better solution an those that currently exist, please follow the instructions to submit your own response.
What's in it for you?
The plan is to submit a peer reviewed manuscript showing the results of these challenges. Those who participate will be listed as co-authors. The winner of the challenge will be the first author. Companies are more than welcome to participate.
If you are a fan of open science, Hadoop, Spark, and other Big Data projects, then this challenge is for you!