HPC Solutions for Life Science Research Glen Otero

HPC Solutions for Life Science Research Glen Otero, Ph. D. Bioinformatics & NGS Solution Architect HPC & Research Computing Glen_Otero@Dell. com

Partnership Accelerating Individual Treatments for Pediatric Cancer Dell HPC

Precision Medicine At Work Today Against “Adult” Cancers • Look inside cancer cells, identifyspecific vulnerabilitiesand customize a treatment • Once the molecular-mechanisms of disease are aligned with the righttargeted drugwe can dramatically improve patientsurvival. • We will do this for children with neuroblastoma Examples of Precision Medicine Leukemia • Gene Target: ECR-ABL • 30% increase in survival rate • 41 yrs from indentified cause to targeted treatment Breast Cancer • Gene Target: HER 2 • 20% increase in survival rate • 17 yrs from identified cause to targeted treatment Melanoma • Gene Target: BRAF • 52% response rate and increased survival by six months • 10 yrs from identified cause to treatment Lung Cancer • Gene Target: EGFR • Partial or complete response in 93% of patients with DNA mutation • 9 yrs from identified cause to treatment Medical imaging proof of dramatic tumor reduction after 8 weeks on personalized treatment regimen. N Engl J Med. 2009 Sep 17; 361(12): 1173 -8. Dell HPC

Why Pediatric Cancer? • Pediatric Oncology – Leading cause of disease-related death in children ages 1 -14 – A child diagnosed with cancer every hour – Cure rates have not improved in the last decade – Pharmaceutical companies fund 60% of adult oncology research and virtually no pediatric cancer research • Neuroblastoma (NB) – Worst clinical outcome of any pediatric cancer – 5% chance of cure in children with advanced cancer – Even with remission, the cancer often returns and is untreatable Dell HPC

Dell & TGen collaboration will result in. . . • More children offered a chance at a cure for cancer • 100% increase in patients treated the first year and a future global platform to treat 100 K+ children and adults Dell HPC

Applying Precision Treatments for Pediatric Cancer Neuroblastoma: 15 days for personalized treatment Patient / Physician diagnosis, treatment, ongoing management Treatments with a more reasonable chance of a cure Tumor Sample Minimizing trial and errors Archival Storage Complete molecular characterization of the diseased tumor Understanding the individual disease Analytical tool for mapping patient data against database for recommended treatment Accelerating targeted treatment options Integration of scientific & clinical evidence for future research Creating platform to scale to 100 k+ patients Analytics HPCC Dell HPC

Genomics Data Processing Pipeline

TGen RNA-Seq. Data Analysis QC Trim Visualize average q-score for each sequencing cycle Trim off bases starting with the under-performing cycles Align to Genome Map Splice Junctions Assemble Isoforms and Estimate Abundances Visualize Regions of Interest Using Bowtie Langmead B, Trapnell C, Pop M, Salzberg SL. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Using Tophat Trapnell C, Pachter L, Salzberg SL. Top. Hat: discovering splice junctions with RNA-Seq. Using Cufflinks Using IGV Trapnell C, Williams BA, Pertea G, Mortazavi AM, Kwan G, van Baren MJ, Salzberg SL, Wold B, Pachter L. Transcript assembly and quantification by RNA-Seq reveals unannotated transcripts and isoform switching during cell differentiation. The Broad Institute, Development: Jim Robinson, Helga Thorvaldsdóttir, Marc-Danie Nazaire, Documentation: Judy Mc. Laughlin, PI: Jill Mesirov Global Marketing

TGen RNA-Seq Performance Data analysis time was 7 days. Goal was 5 days. Worked with TGen to parallelize their RNA-Seq pipeline 9 Confidential Global Marketing

M 420 s Energy Efficiency Application Power. Edge R 820 @ 2. 7 GHz 4 servers, 128 cores Power. Edge M 420 @ 2. 3 GHz 32 servers, 512 cores HPL EE: 4% higher Best EE: 15% higher LU Best EE: ~6% higher Similar EE to Power. Edge M 620 WRF EE: 12% lower Best EE: ~13% higher ANSYS Fluent Best EE for truck_poly_14 m (+16%) Best EE for truck_111 m (+23%) MILC EE: ~18% lower Best EE: ~37% higher NAMD (Issues when executing program) Better EE: ~5 % higher * EE is shorthand for energy efficiency. The Power. Edge M 620 cluster (16 @ 2. 7 GHz) is used as the baseline for comparison. Higher is better for EE. 10 Confidential Global Marketing

Why the M 420 s GFLOPS/$ GFLOPS/U 900 800 700 600 500 400 300 200 100 0 826. 8 0. 080 0. 070 504. 4 0. 060 448. 4 411 0. 040 0. 030 0. 020 0. 010 M 420 (32 @ 2. 3) M 620 (16 @ 2. 7) M 620 (16 @ 2. 4) M 620 (16 @ 2. 2) GFLOPS/W M 420 (32 @ 2. 3) Confidential M 620 (16 @ 2. 7) M 620 (16 @ 2. 4) 0. 000 M 420 (32 @ 2. 3) M 620 (16 @ 2. 7) M 620 (16 @ 2. 4) M 620 (16 @ 2. 2) • • • 0. 900 0. 800 0. 700 0. 600 0. 500 0. 400 0. 300 0. 200 0. 100 0. 000 11 0. 050 48 GB on M 420 and 64 GB on M 620 130, 115 & 95 Watt Processors 20% more GFLOPS/$ with M 620 s 64% more GFLOPS/U with M 420 s 15% more GFLOPS/W with M 420 s M 620 (16 @ 2. 2) Global Marketing

TGen RNA-Seq Performance Data analysis time was 7 days. Goal was 5 days. Worked with TGen to parallelize their RNA-Seq pipeline One 10 U chassis of 32 x M 420 s can complete RNA-Seq pipeline in 4 hours. 12 Confidential Global Marketing

TGen RNA-Seq Performance Data analysis time was 7 days. Goal was 5 days. Worked with TGen to parallelize their RNA-Seq pipeline One 10 U chassis of 32 x M 420 s can complete RNA-Seq pipeline in 4 hours. 5 days hoped for. 4 hours delivered. 13 Confidential Global Marketing

The Dell Scalable Unit (DSU) for Life Sciences: SANGER (Sequence Analysis ‘N’ GEnomics Research) Glen Otero, Ph. D. Mark R. Fernandez, Ph. D. Jeff Layton, Ph. D. Life Sciences HPC Solution Architect HPC Computer Scientist HPC Enterprise Technologist Dell | Research Computing Solutions glen_otero@dell. com mark_fernandez@dell. com jeffrey_layton@dell. com DELL CONFIDENTIAL *** DRAFT*** (Mark R. Fernandez: v 1. 4. 6. 1, 12 -June-2012)

SANGER Infrastructure: Dell PE, PC & F 10 NSS-HA Pair Dell NSS (NFS) (up to 180 TB) NSS User Data HSS Metadata Pair HSS OSS Pair Dell HSS (Lustre) (up to 360 TB) M 420 (Compute) (up to 32 nodes) Actual placement in racks may vary. • Solution: Dell NGS Cluster Appliance – Single Rack Solution – 9 Teraflops of Sandy Bridge Processors – Lustre File Storage – Intel SW tools • Benefits: Includes everything you need for NGS - compute, storage, software, networking, infrastructure, installation, deployment, training, service & support HSS User Data 2 U Plenum • Challenge: Experiment processing takes too long and delays patient treatment

Genomic Research - Customer. Challenges Data explosion • Large data and compute requirements Data management, access, migration are challenging • Processing required measured in days and weeks • Resource requirements inhibit clinical / commercial viability Innovation, iteration and refinement • • Genomic software and pipelines rapidly evolving Rich set of open source, shareware, and commercial applications Sequence Align Identify Variants Visualize Define Therapy Cumbersome infrastructure • • • Costly, complex deployment, overbearing maintenance Shared, limited-access resources Confidentiality and privacy concerns Archive Global Marketing

whole genome pipeline analysis overview Dell HPC DELL CONFIDENTIAL *** DRAFT*** (Mark R. Fernandez: v 1. 4. 6. 1, 12 -June-2012) 17

SANGER whole genome analysis info • A variant calling pipeline on public NA 12878 data from Illumina's Platinum Genome project: http: //www. illumina. com/platinumgenomes • Aligner: bwa; variant caller: GATK • Reference Genome: GRCh 37 (Genome Reference Consortium Human build 37) • http: //www. ncbi. nlm. nih. gov/projects/genome/assembly/grc/ • Software pipeline framework: https: //github. com/chapmanb/bcbio-nextgen • How to install pipeline framework: • https: //bcbio-nextgen. readthedocs. org/en/latest/contents/installation. html • How to run the whole genome pipeline: • https: //bcbio-nextgen. readthedocs. org/en/latest/contents/testing. html#whole-genome • Config file: • https: //github. com/chapmanb/bcbio-nextgen/blob/master/config/examples/NA 12878 illumina. yaml • Input files • ftp. sra. ebi. ac. uk/vol 1/fastq/ERR 091571/ERR 091571_1. fastq. gz • ftp. sra. ebi. ac. uk/vol 1/fastq/ERR 091571/ERR 091571_2. fastq. gz • Number of reads: 212 million (~10 x coverage) • Separate trials conducted by running 1, 6 and 30 genomes analyses simultaneously Dell HPC DELL CONFIDENTIAL *** DRAFT*** (Mark R. Fernandez: v 1. 4. 6. 1, 12 -June-2012) 18

SANGER genomics workload performance 12 human genomes (30 x) processed per day Consume 20. 23 k. Wh per genome Open Integrated 19 10/2/2020 Confidential Efficiency Performance Architected for genomic workloads Service and Support Performance metrics above based on May 2013 internal Dell HPC lab benchmark testing (SW: bwa, GATK, bcbio-nextgen Ref. Gehome: GRCh 37 Input reads: 212 M, ~10 X coverage), confirming 1) 38. 9 genomes per day throughput on an Active Infrastructure for HPC Life Sciences configuration across 480 cores 2) confirming 38. 9 genomes per day throughput compared to 2. 01 genomes per day across 16 cores. 3) the complete analysis of a genome in 6. 8 hours across 96 cores 4) energy consumption of 745. 31 k. W when running 30 concurrent genome analyses, resulting in 24. 85 kilowatts/genome (745. 31 k. W/30 genomes). 5) confirmed sustained 8. 26 Tflops, 88% of a theoretical maximum of 9. 4 Tflops, based on May 2013 internal Dell HPC lab benchmark testing. Actual performance will vary based on configuration, usage and manufacturing variability. Global Marketing

SANGER Realize genomics R&D potential Best-in-class density and energy efficiency Be in production faster Open, factory integrated HPC infrastructure Reduce time to market World class support and expertise Identify treatments in clinically relevant timeframes Leverage Dell domain expertise and complete solutions 20 Confidential 10/2/2020 Innovate Global Marketing

SANGER available in Design Solution Center for testing 21 Confidential 10/2/2020 Global Marketing

Thanks! • Intel • Ketan Paranjape, Kristina Kermanshahche, John O’Neill, Drew Peterson, Ed Kurtzer • Terascala • Ben Rosen, Larry Bazinett • Bright Computing • Matthijs van Leeuwen, Craig Reagan • TGen • James Lowey, Nelson Kick, Jason Corneveaux, Matt Huentelman • Dell • Jeff Layton, Mark Fernandez, Onur Celebioglu, Nishanth Dandapanthula, Will Cottay, Neil Klosterman, Christine Fronczak • Harvard • Brad Chapman, Oliver Hofmann, Rory Kirchner, John Morrissey Dell HPC 22

NGS that drives clinical actions in real time • TGen pediatric cancer trial • http: //www. youtube. com/watch? v=BTMZe 8 EHDPU&feature =bf_prev&list=PLCB 5 D 4 BB 7 F 33065 D 0 • TGen Center for Rare Childhood Disorders • http: //www. youtube. com/watch? v=Ob. Na. YEFKNXo&feature =player_embedded • http: //www. azcentral. com/health/news/20121004 tgenwidening-battle-kids. html • http: //www. news-medical. net/news/20121006/Translational. Genomics-Research-Institute-creates-new-center-for-rarechildhood-disorders. aspx Dell HPC 23

Dell and TGen take the fight against pediatric cancer personally. . . one child at a time Glen Otero, Ph. D. Bioinformatics & NGS Solution Architect HPC & Research Computing Glen_Otero@Dell. com