Big Data Genomics Sequencing
The University of North Carolina at Chapel Hill is using big data to find better treatments for patients through genomic sequencing technologies
Charles Schmitt, Director of Informatics and Data Sciences at RENCI, UNC
The University of North Carolina at Chapel Hill (UNC) decided to invest heavily in genomics sequencing technologies, both to support the clinical-care mission of its healthcare system and to further basic genomic and biology research. This institution-wide initiative is a data-intensive undertaking that requires managing and analyzing hundreds to thousands of individual genomes to fit the different needs of clinicians and researchers.
To address this big data challenge, we use a three-stage process. It starts in the wet lab, where a patient’s tissue is collected. Then a combination of 11 production high-throughput genomics sequencers, made by Illumina and Pacific Biosciences, generate hundreds of millions of short DNA sequences for each patient. With that data, researchers use parallelized computational workflows to reassemble the genome and perform quality control on the reassembly—fixing errors in the reassembly.
Once the genomes are pulled together, our second stage detects variations for an individual, often using large patient populations to help resolve ambiguities in the individual’s sequence data. This data is organized into a hybrid solution that uses a relational database to store canonical variations, high-performance file systems to hold data, and a Hadoop*-based approach for data-intensive analysis. Links to public domain and private databases help researchers identify the impact of variations on protein formation, whether the variants are known to be associated with clinically relevant conditions, and other known characteristics of the variation.
The final step is the report to the physicians—the third stage of the process. Key to this stage is a process termed “clinical binning” that is performed using a custom-made piece of software that UNC has developed. This assigns a clinical relevancy to every one of the variants, allowing clinicians and patients to determine the types of variants they care about. Once variants are binned, we have a web site that delivers the information on the individual to his or her physicians. This stage also manages the overall process, from blood-draw to analysis to reporting, including several stages that provide independent validation of the identified variants.
Our solution relies on a large commodity cluster that uses 50 Intel®-processor-based blade servers processing up to 30 genomes a week. Right now, we have about 200 to 300 terabytes of genomic data on a large EMC Isilon* data system.
The management of genomic data is orchestrated through a UNC-based data grid technology called Enterprise iRODS*. Security for sensitive genomic and patient data is provided by the UNC-built Secure Medical Workspace* technology.
The Hadoop system allows us to perform very custom analysis that you wouldn’t find in a traditional business intelligence tool or that would work in a SQL* relational type of structure. Our analyses amend well to a MapReduce* structure. The other issue is that tests with databases that use extract, transform, load (ETL) take an incredibly long time with that much data. With Hadoop there’s no ETL; we just add a file into the system.
We’ve learned a lot about big data analytics. For example, we made investments in a few technologies that were designed to provide flexibility in running workflows on different computational clusters—before we knew what we really required. Instead, we would have been better off by focusing on tailoring our specific analytics needs for our existing infrastructure. But this is always a challenging balance to make.
The thing we did best was staying agile in our processes and being flexible in our technical approaches. After all, genomics sequencing is such a new technology, and things are changing rapidly. The questions that people are asking today are changing just as rapidly, so the informatics solutions have to be similarly adjustable.