2011 Big Data - Bigger Problems for Drug Discovery and Development
1 like866 views
This document summarizes a presentation about using topological data analysis to solve problems in drug discovery and development. It discusses the challenges of big data in this field, such as scalability issues and a lack of integrated tools. It then describes how the Ayasdi Iris platform uses principles of topology to identify subtle patterns in large, complex datasets. Examples are given of how Ayasdi has used topological data analysis for patient stratification and to analyze next-generation sequencing data.
2011 Big Data - Bigger Problems for Drug Discovery and Development
- 1. Bio-IT Pek Yee Lum, Ph.D. VP Life Sciences 1
- 2. Today’s Topics What are the problems we face with big data today in the drug discovery and development world? What problems can Ayasdi solve for you? Why topology? Patient stratification using topology Unveiling our NGS pipeline and genome browser Summary © 2012 Ayasdi inc. 2
- 3. Today’s Topics What are the problems we face with big data today in the drug discovery and development world? What problems can Ayasdi solve for you? Why topology? Patient stratification using topology Unveiling our NGS pipeline and genome browser Summary © 2012 Ayasdi inc. 3
- 4. Big data, bigger problems for drug discovery and development • Ever-growing complex and disparate datasets • Scalability issues • NGS raw data sometimes as large as 1TB per sample • Accessing data no longer simple for the untrained users • New IT infrastructure for every new problem • Every new data type needs custom tools • User experience does not exist today • Bioinformatics tools are not integrated • Analysis and visualization is disparate • Accelerating the discovery process requires rethinking the analysis workflow and streamlining its computational infrastructure Problems in drug discovery and development © 2012 Ayasdi inc. 4
- 5. Data analysis landscape today R Cytoscape Database Spotfire Matlab Math Writing code Cloud Problems in drug discovery and development © 2012 Ayasdi inc. 5
- 6. Biological Complexity on top of data problems Diseases are often complex. Many components work in synergy for disease manifestation. The need to identify perhaps not a single drug target but multiple targets that work as a network Human population is heterogenous- drugs fail because of the inability to stratify the patient population The need to identify biomarkers that work for patient stratification to decrease risk of adverse events and lack of efficacy Problems in drug discovery and development © 2012 Ayasdi inc. 6
- 7. Today’s Topics What are the problems we face with big data today in the drug discovery and development world? What problems can Ayasdi solve for you? Why topology? Patient stratification using topology Unveiling our NGS pipeline and genome browser Summary © 2012 Ayasdi inc. 7
- 8. Ayasdi Iris increases probability of success (POS) and shrinks time to market Discovery of subtle patterns in a sea of noisy data Handling of all data- large or small on the cloud Fusing disparate data sets with ease Access to critical public data on demand Allows collaboration for all types of stakeholders on one platform The Ayasdi solution © 2012 Ayasdi inc. 8
- 9. Today’s Topics What are the problems we face with big data today in the drug discovery and development world? What problems can Ayasdi solve for you? Why topology? Patient stratification using topology Unveiling our NGS pipeline and genome browser Summary © 2012 Ayasdi inc. 9
- 10. Solution Math+CS+UX platform Data has Shape and Shape has Meaning Data has shape © 2012 Ayasdi inc. 10
- 11. What is shape ? If Age, Weight and Height were In reality, age, weight and height are distributed randomly correlated and that data has a shape Data has shape © 2012 Ayasdi inc. 11
- 12. What is shape ? Ayasdi Iris identifies the shape or pattern in data Data has shape © 2012 Ayasdi inc. 12
- 13. Why Topological Data Analysis for drug discovery and development 1. Coordinate free representations are vital when studying data collected using different technologies- lots of public data available, many studies done at different times, different data types collected 2. Deformation invariance introduces robustness into the analysis, which is important in the study of real world data- biological heterogeneity is complex and needs an approach that is deformation (variation) resistant 3. Compressed representations are obviously important when one is dealing with very large data sets- with high dimensional omics data and Next Gen sequencing getting more affordable, the amount of data is increasing exponentially Why topology ? © 2012 Ayasdi inc. 13
- 14. Ayasdi Iris Uses principles of geometry to find shape (pattern) in data Works across and for any type of data Works with any amount of data Generates and validates hypotheses Quick, interactive results Why topology ? © 2012 Ayasdi inc. 14
- 15. Patient stratification - Results gene expression profiling of breast tumors Identified a sub-group of patients that are triple negative with very good prognosis Identified a sub-group of patients that are Luminal A but with perfect survival (published PNAS 2011) These groups were identified in independent datasets These sub-groups were hard to find using conventional methods Patient stratification © 2012 Ayasdi inc. 15
- 16. Today’s Topics What are the problems we face with big data today in the drug discovery and development world? What problems can Ayasdi solve for you? Why topology? Patient stratification using topology Unveiling our NGS pipeline and genome browser Summary © 2012 Ayasdi inc. 16
- 17. Each node contains subsets of patients These patients are eccentric (away from the center of the data) These patients are close to the center of the data Color scheme Topological Map of Patient-Patient Relationships according to the molecular characteristics of their tumors (in this case, gene expression) Patient stratification © 2012 Ayasdi inc. 17
- 18. Zero event death Low event death High event death High event death Low event death Mixed event death These patients are eccentric (away from the center of the data) Zero event death Color scheme Color scheme Patient stratification Zero event death © 2012 Ayasdi inc. 18
- 19. patients did not survive 10 yrs B D E A Triple Negative HER2-, ESR-, PGR- C patients survived 10 yrs Patient stratification © 2012 Ayasdi inc. 19
- 20. NKI data high ESR1 death low ESR1 survived high ESR1 low ESR1 Topological maps from GSE2304 two independent cancer data sets are very relapsed low ESR1 similar high ESR1 relapse no high ESR1 low ESR1 © 2012 Ayasdi inc. 20
- 21. Conventional Methods subtypes difficult to identify Clustering PCA ER- did not survive ER- survived Patient stratification © 2012 Ayasdi inc. 21
- 22. Next Generation Sequencing unveiling our solution DNA-Seq Pipeline BAMs VCF IRIS RNA-Seq Pipeline BAMs cuffcomp/RPKM IRIS 22
- 23. 1000 genomes DNA-seq data population classification Africans East Asians Caucasians 23
- 24. Identification and visualization of significant exon variants using PCA Han Chinese and Japanese cannot be easily distinguished using PCA Han Chinese (grey) Japanese (red) 24
- 25. Identification and visualization of significant variants Han Chinese and Japanese can be easily distinguished and visualized with TDA Han Chinese rs2294008 TT chr8 Japanese Poster #3: Navigating Next Generation Sequencing Data using Topological Data Analysis and Iris 25
- 26. Today’s Agenda What are the problems we face in drug discovery and development? What problems can Ayasdi solve for you? Why topology? Patient stratification using topology Summary © 2012 Ayasdi inc. 26
- 27. Ayasdi Iris platform Ayasdi Iris Life Sciences Edition All analyses performed on the cloud on secure servers PubMed GO Bypass the need to invest in expensive hardware and KEGG PPI database administration GEO Interactive Network Visualization Integrated Statistics Algorithms Integrated Public Datasets Flexibility to start your analysis immediately- just upload your data Ayasdi Iris Cloud Platform Access to public data on demand Analysis and Visualization Topological Network Projections Network Histograms/ Dendro- Mapping Analysis (e.g. PCA) Visualization Scatterplots grams Scalable Distributed Datastore Proprietary and Public data sources Gene mRNA SNP Clinical NGS PubMed Expression Ayasdi Iris platform © 2012 Ayasdi inc. 27
- 28. Ayasdi Iris increases probability of success (POS) and shrinks time to market Discovery of subtle patterns in a sea of noisy data Handling of all data- large or small on the cloud Fusing disparate data sets with ease Access to critical public data on demand Allows collaboration for all types of stakeholders on one platform The Ayasdi solution © 2012 Ayasdi inc. 28
- 29. Contact us for more information www.ayasdi.com pek@ayasdi.com info@ayasdi.com 29