SUMMARY Genomic information is increasingly incorporated into medical practice for diagnosis and personalised treatment. However, processing genomic information at a scale relevant for the health-system remains challenging due to computational requirements as well as high demands on data reproducibility and data provenance. Here, we present Next Generation Sequencing Analysis for Enterprises (NGSANE), a Linux-based, High Performance Computing (HPC) framework for production informatics, tailored to the demands and fast pace of personalised medicine, which is available as on-demand virtual cluster in Amazon's Elastic cloud.
Individual task blocks (e.g. read mapping) are packaged into bash script modules, which can be executed locally or on data subsets to test module code, submission parameters and compute environment in stages thereby mitigating the lack of debug-support from higher level languages/submission frameworks. During production, NGSANE automatically submits separate module calls for each individual data set to the HPC queue. This allows different existing modules, parameter settings, or software versions to be executed by changes to the project specific configuration file rather than the software code (hot swapping). NGSANE gracefully recovers from unsuccessfully executed jobs be it due to failed commands, missing or incorrect input or under-resourced HPC jobs by enabling a clean restart from the most recent successfully executed checkpoint. Workflows can be fully automated by utilising NGSANE’s control over HPC queuing systems and by leveraging the customisable interfaces between modules when submitting multiple dependent stages at once.
NGSANE supports the generation of a high-level summary (Project Card) to enable informed decisions about the experimental success. This interactive HTML report provides an access point for new lab members or collaborators, as well as a gold standard that can be used for testing purposes in a continuous integration server framework.
NGSANE is available as an Amazon Machine Image (AMI), which can be deployed to Amazon’s EC2 by using, for example, MIT’s StarCluster framework (http:// star.mit.edu/cluster/) to launch a virtual cluster on demand (see Figure 1B). Other than regular on-demand instances, whose availability is guaranteed at a fixed price, StarCluster also offers command line-based sourcing of Spot Instances, where prices are based on current supply and demand. While Spot Instances can be acquired at a substantially lower price, their availability is not guaranteed. Hence NGSANE’s checkpoint recovery is critical in such an unstable, competitive environment. Finally, NGSANE’s HPC job partitioning and submission structure is independent from the program calls, therefore allowing new technologies (e.g. Hadoop) to be incorporated.
1. A) Resource consumption of the four steps involved in exon capture genomic data analysis. The average per sample is plotted in hours and gigabytes for CPU usage (single and multithreaded) and RAM memory usage, respectively. B) Schematic for a nine-node on-demand cluster with the NGSANE AMI deployed on every node on the EC2 service as launched by StarCluster.
CONCLUSION NGSANE is a flexible HPC framework for NGS data analysis that is specifically tailored to the demands and issues of personalised genomics. NGSANE is implemented in bash and publicly available under BSD (3-Clause) licence via GitHub at https://github.com/BauerLab/ngsane. Currently implemented workflows include those for adapter trimming, read mapping, peak calling, motif discovery, transcript assembly, variant calling and chromatin conformation analysis. NGSANE is available for local cluster installation or as an AMI to be deployed as an on-demand cluster on Amazon’s EC2. This facilitates production-scale processing of large sample numbers and enables research at population scale to produce insights into individual disease risk and stratify treatment for common diseases with impact on the health system.