CLOUD BIOINFORMATICS Part1
Download as PPT, PDF0 likes917 views
This document discusses using cloud computing to address challenges in genome informatics posed by exponentially growing genomic data. It outlines how the traditional ecosystem is threatened as DNA sequencing costs decrease faster than storage and computing capacity can grow. Cloud computing provides an alternative by allowing users to rent vast computing resources on demand. The document examines applying MapReduce frameworks like Hadoop and DryadLINQ to bioinformatics applications like EST assembly and Alu clustering. Experiments showed these approaches can simplify processing large genomic datasets with performance comparable to local clusters, though virtual machines introduce around 20% overhead. Overall cloud computing may become preferred for its flexibility and ability to move computation to data.
CLOUD BIOINFORMATICS Part1
- 1. GENOME CLOUD COMPUTING By.A.Arputha Selvaraj
- 2. Introduction The impending collapse of the genome informatics Major sequence-data relateddatabases: GenBank, EMBL, DDBJ, SRA, GEO, and ArrayExpress Value-added integrators of genomic data: Ensembl, UCSC Genome Browser, Galaxy, and many model organism databases.
- 3. The old genome informatics ecosystem
- 4. Basis for the production and consumption of genomic information Moore's Law: “the number of transistors that can be placed on an integrated circuit board is increasing exponentially, with a doubling time of roughly 18 months.” Similar laws for disk storage and network capacity have also been observed. Until now the doubling time for DNA sequencing was just a bit slower than the growth of compute and storage capacity.
- 5. Historical trends in storage prices vs DNA sequencing costs Notice: this is a logarithmic plot – exponential curves appear as straight lines. “Next generation” sequencing technologies in the mid-2000s changed the trends and now threatens the conventional genome informatics ecosystem
- 6. Current Situation Soon it will cost less to sequence a base of DNA than to store it on hard disk We are facing potential tsunami of genome data that will swamp our storage system and crush our compute clusters Who is affected: Genome sequence repositories who need maintain and timely update their data “Power users” who accustomed to download the data to local computer for analysis
- 7. Cloud Computing Cloud computing is a general term for computation-as-a-service Computation-as-a-service means that customers rent the hardware and the storage only for the time needed to achieve their goals In cloud computing the rentals are virtual The service provider implements the infrastructure to give users the ability to create, upload, and launch virtual machines on their extremely large and powerful compute farm.
- 8. Virtual Machine Virtual machine is software that emulates the properties of a computer. Any operating system can be chosen to run on the virtual machine and all the tasks we usually perform on the computer can be performed on virtual machine. Virtual machine bootable disk can be stored as an image. This image can used as a template to start up multiple virtual machines – launching a virtual compute cluster is very easy. Storing large data sets is usually done with virtual disks. Virtual disk image can be attached to virtual machine as local or shared hard drive.
- 9. The “new” genome informatics ecosystem based on cloud computing Continue to work with the data via web- pages in their accustomed way Have two options: continue downloading data to local clusters as before, or use cloud computing – move the cluster to the data.
- 10. Cloud computing service providers and their offered services Some of the cloud computing service providers are: Amazon (Elastic Cloud Computing), Rackspace Cloud, Flexiant. Amazon offers a variety of bioinformatics-oriented virtual machine images: Images prepopulated by Galaxy Bioconductor – programming environment with R statistic package GBrowser – genome browser BioPerl JCVI Cloud BioLinux – collection of bioinformatics tools including Celera Assembler Amazon also provides several large genomic datasets in its cloud: Complete copy of GeneBank (200 Gb) 30x coverage sequencing reads of a trio of individuals from 1000 Genome Project (700Gb) Genome databases from Ensembl including annotated human and 50 other species genomes
- 11. Cloud computing service providers from Nature Biotechnology news article “Up in a cloud?”
- 12. Academic cloud computing Growing number of academic compute cloud projects are based on open source cloud management software. Open Cloud Consortium is an example of one such project
- 13. Cloud computing disadvantages Cost of migrating existing systems into the new and unfamiliar environment Security and privacy of the data Need additional level of data encryption Need new software that will restrict access to only authorized users Major question: does cloud computing make economic sense?
- 14. Comparing relative costs of renting vs. buying computational services According to technical report on Cloud Computing by UC Berkeley Reliable Adaptive Distributed Systems Laboratory When all the costs of maintaining a large compute cluster are considered the cost of renting a data center from Amazon is marginally more expensive than buying one. When the flexibility of the cloud to support a virtual data center that downsizes or grows as needed is factored in, the economics start to look better
- 15. Obstacles in moving toward cloud computing for genomics Network bandwidth – transferring 100 gigabit next-generation sequencing data file across 10Gb/sec connection will take less than a day only if hogging much of the institution’s bandwidth. Cloud computing service providers will have to offer some flexibility in how large datasets get into system.
- 16. By Jaliya Ekanayake, Thilina Gunarathne, and Judy Qui IEEE transactions on parallel and distributed systems Cloud Technologies for Bioinformatics Applications
- 17. Motivation for presenting the article Exponential increase in the size of datasets implies that parallelism is essential to process the information in a timely fashion. The algorithms presented in this paper might reduce computational analysis time significantly whether used on our own cluster or on rented Amazon virtual cluster. This paper presents a nice example of how these new parallel computing algorithms can be applied for bioinformatics
- 18. MapReduce MapReduce is a software framework introduced by Google to support distributed computing on large data sets on computer clusters. Users specify a map function that processes a key/value pair to generate a set of intermediate key/value pairs, and a reduce function that merges all intermediate values associated with the same intermediate key. Programs written in this functional style are automatically parallelized and executed on a large cluster of commodity machines. The run-time system takes care of the details of partitioning the input data, scheduling the program's execution across a set of machines, handling machine failures, and managing the required inter-machine communication. This allows programmers without any experience with parallel and distributed systems to easily utilize the resources of a large distributed system.
- 19. Tow steps of MapReduce Map: The master node takes the input, chops it up into smaller sub-problems, and distributes those to worker nodes. The worker node processes that smaller problem, and passes the answer back to its master node. Reduce: The master node then takes the answers to all the sub-problems and combines them in a way to get the output - the answer to the problem it was originally trying to solve.
- 21. Objectives Examine 3 technologies: Microsoft Drayd/DryadLINQ Apache Hadoop Message Passing Interface (MPI) These 3 technologies are applied to 2 bioinformatics applications: Expressed Sequence Tags (EST) Alu clustering
- 22. Technolgies MPI – The goal of MPI is to provide a widely used standard for writing message passing programs. MPI is used in distributed and parallel computing. Dryad/DryadLINQ is a distributed execution engine for coarse grain data parallel applications. It combines the MapReduce programming style with dataflow graphs to solve computational problems. Apache Hadoop is an open source Java implementation of MapReduce. Hadoop schedules the MapReduce computation tasks depending on the data locality.
- 23. Bioinformatics applications Alu Sequence Classification is one of the most challenging problems for sequence clustering because Alus represent the largest repeat families in human genome. For the specific objectives of the paper an open source version of the Smith Waterman – Gotoh (SW-G) algorithm was used. EST (Expressed Sequence Tag) corresponds to messenger RNAs transcribed from genes residing on chromosomes. Each individual EST sequence represents a fragment of mRNA, and the EST assembly aims to reconstruct full-length mRNA sequence for each expressed gene. Specific implementation used for this research is CAP3 software.
- 24. Scalability of SW-G implementations
- 25. Scalability of CAP3 implementations
- 26. How different technologies perform on cloud computing clusters The authors were interested to study the overhead of Virtual Machines for application that use frameworks like Hadoop and Dryad. Performance degradation of Hadoop SWG application on virtual environment ranges from 15% to 25%
- 27. How different technologies perform on cloud computing clusters cont’d The performance degradation in CAP3 remains constant near 20% . CAP3 application does not show the decrease of VM overhead with the increase of problem size as with SWG application.
- 28. Conclusions The flexibility of clouds and MapReduce suggest they will become the preferred approaches. The support for handling large data sets, the concept of moving computation to data, and the better quality of service provided by the cloud technologies simplify the implementation of some problems over traditional systems. Experiments showed that DryadLINQ and Hadoop get similar performance and that virtual machines five overheads of around 20%. The support for inhomogeneous data is important and currently Hadoop performs better than DryadLINQ. MapReduce offer higher level interface and the user needs less explicit control of the parallelism.
- 29. Thank you