MawereC- Ubuntunet paper publication 2015

Page | 1 ISSN 2223-7062 Proceedings and report of the 7th UbuntuNet Alliance annual conference, 2014,
pp 217-230
Cloud-Based Big Data Analytics in
Bioinformatics: A Review
Cephas MAWERE1
, Kudakwashe ZVAREVASHE2
, Thamari SENGUDZWA3
, Tendai
PADENGA4
1
Harare Institute of Technology, School of Industrial Sciences & Technology Biotech. Dept.,
P.O. Box BE277 Belvedere, Harare, Zimbabwe
Tel: +263 772 384985, +263 4 741 422-58, Fax: +263 4 741 406
Email:cmawere@gmail.com
2
Harare Institute of Technology, School of Information Science IT Dept., P.O. Box BE277
Belvedere, Harare, Zimbabwe
Tel: +263 733 784352, +263 4 741 422-36, Fax: +263 4 741 406
Email:kzvare@gmail.com
3
Harare Institute of Technology, School of Industrial Sciences & Technology Biotech. Dept.,
P.O. Box BE277 Belvedere, Harare, Zimbabwe
Tel: +263 774 088979, +263 4 741 422-57, Fax: +263 4 741 406
Email: thamaryc@gmail.com
4
Harare Institute of Technology, Dean of School of Information Science, P.O. Box BE277
Belvedere, Harare, Zimbabwe
Tel: +263 775 958368, +263 4 741 422-36, Fax: +263 4 741 406
Email: tepadenga@gmail.com
Abstract
The significant advances in high- throughput sequencing technologies over the last decade
have led to an exponential explosion of biological data. Consequently, bioinformatics is
encountering unprecedented challenges in data storage and analysis, memory allocation,
computational power and time complexity. It is therefore becoming increasingly disturbing
for small laboratories and some large institutions to establish and maintain infrastructures for
data processing. This has forced bioinformatics to take a leap-forward from in- house
computing to cloud computing to address these issues. In this paper, the authors thus explore
the current applications of big data analytics in bioinformatics on the cloud. The findings in
each in each use case reveal that cloud computing remarkably improve processing and
storage of the continuously growing data in bioinformatics. Though there are some
bottlenecks to solve, cloud computing promises to provide a lightweight environment to
develop pipelines for prognosis, diagnosis, drug discovery and personalized medicine. The
knowledge generated by this review will help scientist who are facing challenges with big
data to resort to use cloud computing as an alternative solution and get results faster and
without the need to install and maintain or update expensive software. This paper also serves
to bring awareness to scientists on the current technologies that are used to manage data.
Keywords: High-throughput Sequencing Technologies; Bioinformatics; Data Processing;
In- House Computing; Cloud Computing; Big Data Analytics.

pp 217-230
1. Introduction
In the last seven years, more scientific data has been generated than in the entire human
history. Hide, W. (2012). This has been triggered by the recent advances in high throughput
technologies such as Next Generation Sequencing (NGS), genotyping, Single Nucleotide
Polymorphism (SNP) discovery, gene expression, proteomics as well as virtual screening
Blankenburg, L, et al (2009). For example, GenBank has seen DNA sequencing data
doubling every 14 months in the last decade as attested by Figures 1-3. As on 21 October
2014, sequence data has been recorded to have reached 108
bytes while the number of bases
has catapulted to 1011
and 1012
bytes for GenBank WGS data respectively GeneBank. (2014).
Figure 1: Sequence data from GenBank statistics [3]

pp 217-230
Figure 2: GenBank statistics on number of bases [3]
Figure 3: Whole Genome Shotgun (WGS) base data statistics [3]
Inevitably, biological data has exploded beyond the capacity of computers to capture, store,
analyse, search, share, visualize and transfer it for processing in the required time (Monica,
N. & Kumar, R. (2013), Such big data has daunted small laboratories, hospitals and even
some large institutions to establish and maintain infrastructures for data processing Dai, L., et
al (2012). This has forced bioinformatics to leap forward from in-house computing to cloud
computing which promises to address these issues.
But what is big data? Well, the term big data is defined as “a collection of data sets so large,
varied, frequently updated and complex that it becomes difficult to process using on-hand
data management tools” Movirtu. (2014a) Movirtu. (2014b). The process of research into such data
to reveal hidden patterns and secret correlations is thus called ‘big data analytics’
Implementa. (2014). This research therefore seeks to explore the current applications of big
data analytics in bioinformatics on the cloud. In each of the applications the authors will
highlight how cloud computing has improved the processing of such exponentially growing
biological data.
The paper will thus be composed of 5 categories. Section 2 will look at the overview of cloud
computing and its application in bioinformatics. Consequently section 3 elaborates the
applications of bioinformatics on the cloud for each area of high throughput technology and
section 4 brings out the advantages and challenges of big data analytics on the cloud. Finally
the conclusion and recommendation sums it up in section 5 and 6 respectively.
2. Cloud Computing In Bioinformatics
2.1 Overview of cloud computing
The buzzword ‘cloud computing’ was inspired by the cloud symbol that is often used in
reference to the Internet in flowcharts. However, cloud computing is a form of computation

pp 217-230
that exploits the full potential of multiple computers and delivers dynamically allocated
virtual resources via the internet Telco Review (2014). These hosted resources include storage,
computation, applications, servers and network. Such services can be accessed on-demand (in
a pay-as-you-go model) by any user via Web Application Programming Interfaces (API),
without the need of the user knowing where the services are hosted or how they are delivered.
Such virtualization of data in the cloud enables ‘snapshots’ of large data sets to be moved
from point to point within the cloud at high transfer rates without associating particular
machines or storage devices at either source or destination of the data transfer Butte, A. J. &
Dudley, J. T. (2010). This has necessitated Internet-based companies like Google, Windows and
Amazon, whose cloud architectures can harness petabyte scales of data, to offer on-demand
services to tens of thousands of users simultaneously Lin, Y et al (2013).
2.2 Bioinformatics resources on the cloud
Bioinformatics clouds thus involve a large variety of services for big data analytics.
Generally, the clouds fall into four categories (Figure 4) which are Data as a Service (DaaS),
Software as a Service (SaaS), Platform as a Service (PaaS) and Infrastructure as a Service
(IaaS) Stanoevska-Slabeva, K. & Wozniak, T. (2010). DaaS enables dynamic data access on
demand and provides up-to-date that are accessible by a wide range of devices that are
connected over the Web. For instance, Amazon Web Services (AWS) which provides a
centralized repository of public data sets like 1000 Genomes.
Figure 4: Illustration of bioinformatics resources on the cloud Dai, L (2012), Stanoevska-Slabeva, K. &
Wozniak, T. (2010)
IaaS offers full computer infrastructure in form of virtualized resources like CPUs and
operating systems via the Internet. On the other hand, PaaS provides an environment for users
to develop, test and deploy cloud applications where computer resources automatically scale

pp 217-230
to match application demand. The platforms commonly used thus include the Amazon Elastic
Compute Cloud (EC2), Windows Azure and the Google app engine. This research however
focuses on the use of SaaS which delivers software services online and facilitates remote
access to available bioinformatics software tools via the Internet.
3. Applications of bioinformatics on the cloud
In particular, this paper reviews the use of cloud computing to address big data generated in
bioinformatics areas such as (i) sequence mapping, (ii) sequence alignment, (iii) sequence
analysis, (iv) peak calling for ChIP-seq data, (v) identification of epistatic interactions of
SNPs as well as (vi) drug discovery. These applications are run on Amazon and Windows
Azure platforms, with a parallelization implementation of Hadoop’s MapReduce as
illustrated by the example of CrossBow in Figure 5.
3.1 CrossBow- for SNP calling and read mapping
Figure 5: Calling SNPs with CrossBow Russell, J. (2012)

pp 217-230
The MapReduce on Crossbow thus includes the Map step, the Sorting step and the Reduce
step. The Map step uses the Bowtie algorithm (named after the Burrows Wheeler Transform,
BWT) to map sequence reads (which are short DNA sequences about 25-250 base pairs (bp)
long) in parallel to the reference genome. The sequence alignments generated are then
aggregated so that all alignments on the same chromosome or locus are grouped together and
sorted by position. The sorted alignments are then scanned to identify SNPs within each
region using the SOAPsnp (Short Oligonucleotide Analysis Package for SNP) algorithm. The
results are stored via the Hadoop Distributed File System (HDFS), and then archived in
SOAPsnp format Russell, J. (2012)
In the benchmark set on Amazon EC2 cloud, 2.7 billion reads were genotyped in less than
3hrs using a 320 CPU cluster for a cost of $85. Below is a snapshot of Crossbow on the cloud
showing how one can genotype or call SNPs from a genome.
Figure 6: A snapshot of CrossBow on the Amazon EC2 (Gurtowski, J., et al, 2013). On this graphical user
interface (GUI) any user can have his job processed on a pay-as-you go service by specifying the processing
details in the textboxes.

pp 217-230
3.2 CloudBurst
Cloudburst is a new highly sensitive parallel read- mapping algorithm optimized for mapping
next-generation sequence data to reference genomes. Also, it uses the open-source Hadoop
implementation of MapReduce to parallelize execution using multiple compute nodes (Schatz,
M. C., 2009). In a case study done by Schatz in 2008, it was noted that CloudBurst scales
linearly as the number of reads increases, and with near linear parallel speedup as the size of
the number of processors increases (Schatz, M. C., 2009). In a 24- processor core configuration,
CloudBurst is up to 30 times faster than RMAP executing on a single core, while computing
an identical set of alignments.
Figure 7: Running time on EC2 High-Medium Instance Cluster. Comparison of CloudBurst running time
while scaling the size of the cluster for mapping 7million reads to human chromosome 22 with at most four
mismatches on the EC2 Cluster. The 96- core cluster is 3.5X faster than the 24- core cluster Schatz, M. C.
(2009)
Using a larger remote EC2 with 96 cores, CloudBurst improved performance by more than
100 fold. The running time was also reduced from hours to mere minutes for typical jobs
involving mapping of millions of short reads to the human genome as shown by Figure 7.
3.2.1 Genome-Wide Epistatic Interactions
Following high-throughput SNP genotyping, the complex relationship between genotype and
phenotype can now be unravelled on the cloud. As the cardinality of SNPs combinations gets
larger, a dynamic clustering approach is now used for detecting high- order genome- wide
epistatic interactions. In one case study, the speed-up evaluation of the clustering method
(DCHE) was performed on Windows Azure platform on datasets with different sizes ranging
from 10 000 to 100 000. The number of nodes varied from 1 to 40 with 5 minor units (Figure
8).

pp 217-230
Figure 8: Genome-Wide Epistatic Interactions. Computing nodes are sampled from 1-40 with 5 as interval.
The red, blue, cyan and grey curve show functions of speed-up of M=10 000, M= 50 000, and M= 100 000, with
size fixed to 800 Guo, X 2014).
The results showed that when there are a limited number of SNPs in the dataset, for instance,
M= 10 000, it has a lower speed-up curve. This is because dataset with such size along SNP
dimension can be easily processed by a stand-alone version program alone in a couple of
minutes (Guo, X, 2014). However, as the size of datasets increases, speed-up performs better
(green and blue curves). Therefore, the cloud implementation of DCHE has a very good
performance with respect to the speed-up.
3.3 Virtual Screening
Virtual Screening is an established time and cost- effective alternative to in vitro
experiments to enable initial hit finding in the early stage of drug discovery (Certara. 2013). A
virtual screening experiment using Surflex- Dock has been performed in the Amazon
Compute Cloud to find lead structures that bind to the target under investigation. It consisted
of 375 000 ligand-receptor dockings which were then run on 160 cores. Accordingly, the
experiment was completed in < 9hrs for a total of less than $20.

pp 217-230
Figure 9: Surflex-Dock on Amazon EC2.One C1.xlarge EC2= modern computer with 4 cores (-
90hrs).
An EC2 cluster of 40 c1.xlarge instances will thus perform in 2 hours Certara. (2013)
Prior to this, a benchmark study for docking 100 000 structures into a single receptor
conformation was done (Figure 9). It was realised that a single c1.xlarge EC2 instance with
its eight cores delivers nearly the same performance as a reasonably modern desktop
computer with four cores (-90 hours), even though equipped with less CPU power per core.
An EC2 cluster utilizing 20 and 40 c1.xlarge instances is able to perform the same task in 4
and 2 hours, respectively (Certara, 2013).
3.4 Peak calling
Chromatin immunoprecipitation (ChIP), coupled with massively parallel short-read
sequencing (seq) is used to probe chromatin dynamics (Feng, X.et al ,2011.PeakRanger is an
algorithm that is used to resolve closely-spaced peaks on both punctuate sites, such as
transcription factor binding sites, and broad regions like histone modification marks.
In a use case carried out by Feng et. al., (2011) the MapReduce version of PeakRanger
demonstrated that on a fixed number of nodes with increasing data set sizes, the execution
time was shorter and increased more slowly, than the regular single- processor version. For
example, the cloud version processed 14 Gb dataset of 192 million reads in less than 5
minutes, more than 10 times faster than the original PeakRanger (Figure 10a).

pp 217-230
Figure 10: Performance of PeakRanger in cloud parallel computing. A) test with fixed number of
nodes and data sets of increasing size; B)test with increasing numbers of nodes and data sets with fixed
sizes (Feng, X., et al 2011)
In a second test, the scalability of time against the number of nodes was performed. As
expected, the runtime decreased rapidly until the number of nodes equals the number of
chromosomes. Further addition of nodes did not provide further benefit (Feng, X., et al
2011). Thus, PeakRanger on the cloud is provides marked improvements in run time and is
therefore suitable for high-throughput environments.
4. Advantages and Challenges of Cloud Computing
4.1 Benefits of using bioinformatics on the cloud
From the use cases just explored on cloud implementation of bioinformatics tools, cloud
computing has the following advantages:
 It provides a unified, location- independent platform for data & computation. Thus, it
is available even to small labs which can harness its power for big data analytics in
bioinformatics.
 It uses the ‘Pay-per-use’ model which gives access to any user.
 It is flexible; virtual resources can be shared.
 The costs incurred are affordable and relatively cheap for processing of big bio data
 It provides linear scaling of software tools with increasing number of nodes
 It harnesses the power of parallel execution provided by MapReduce
 It is time effective; execution occurs faster than on a local in- house cluster
 It eliminates the need for local installation and eases software maintenances and
updates, providing up-to-date cloud based services.
4.2 Bottlenecks of bioinformatics on the cloud
Nevertheless, the following are challenges commonly faced on the cloud for big data
analytics:

pp 217-230
 Security: - trustability issues. How confidential is the data? There is also limited
control over remote storage. Security updates are installed by vendors and the data is
stored in a redundant manner. Data can be encrypted to secure it as this ensures no
eavesdropping during data transmission (Forer, L et al, 2012).
 Data transfer: Large data sets need to be moved to the cloud. However there is slow
upload speed because of low bandwidth. Some vendors thus have offered to ship data
on hard drives (Forer, L et al, 2012). A promising solution is the use of high speed
transfer technologies such as data compression, Peer-to-Peer data distribution , as well
as Aspera’s fasp high speed file transfer technology (which speeds file transfers over
the Web and outperforms traditional technologies such as FTP and HTTP)
[http://www.aspeerasoft.com/en/technology/fasp_overview_1/fasp_technology_overv
iew_1].
 Scalability: - need for parallelization, and depends on the algorithm itself if
MapReduce suits the use case & desired effects.
 Usability: - bioinformatics involves a pipeline of scripts and a command line
interface. Setting up nodes in the cloud involves a lot of command line and some
deeper understanding. Also adapting new applications to the cloud still requires some
technical expertise.
5. Conclusion
The significant advances in high throughput technologies like NGS have led to an explosion
of biological data. This has given rise to bottlenecks in storage and processing of this data.
This has forced bioinformatics to shift its focus from in- house computing to cloud
computing to address these big data challenges. Here, SaaS was the primary focus of our
review with emphasis on software that can be used on the cloud to address big data
challenges in bioinformatics. Cloud computing of big data in bioinformatics can be done on
several platforms such as Amazon EC2, Windows Azure and Google app engine. Most of
these platforms implement the open source MapReduce software for parallel execution which
makes the cloud preferred over in- house computing in addition to its benefits. Based on
finding obtained from the explored use cases, the cloud- based resources make the big data
meaningful and usable upon processing.
6. Recommendations and Future Works
High speed transfer technologies can be used in future to aid big data transfer. Despite the
challenges existing in cloud computing, future efforts must also be devoted to open and
publicly accessible bioinformatics clouds to the whole scientific community to accelerate
diagnosis, prognosis, drug discovery and personalized medicine. Bioinformatics clouds
should integrate both data and software tools and a lightweight programming environment
should be provided to help in development of pipelines for data analysis.
References
Blankenburg, L., Haberland, L., Elvers, H-D., Tannert, C. & Jandrig, B. (2009) ‘High-
Throughput Omics Technologies: Potential Tools for the Investigation of Influences of EMF
on Biological Systems’ Curr Genomics. 10 (2) pp. 86-92.
Butte, A. J. & Dudley, J. T. (2010) ‘Ín silico Research in the Era of Cloud Computing’
Nature Biotechnology. 28 (11) pp. 1181-1185.

pp 217-230
Certara. (2013) ‘Cloud Computing Enables Cost Effective Virtual Screening’ pp. 1-4.
http://www.certara.com/ [accessed 29 August 2014].
Dai, L., Xiao, J., Zhang, Z., Gao, X. & Guo, Y. (2012) ‘Bioinformatics Clouds for Big Data
Manipulation’ Biology Direct. 7 (43) pp. 1-7.
Feng, X., Stein, L. & Grossman, R. (2011) ‘PeakRanger: A Cloud- Enabled Peak Caller for
ChIP-seq Data’ BMC Bioinformatics. 12 (139) pp. 1- 11.
Forer, L., Schonherr, S., WeinBensteiner, H., et. al. (2012) ‘Cloud Computing’
Computational Medicine. pp. 27-36.
GeneBank. (2014) ‘GeneBank Statistics’http://www.ncbi.nlm.nih.gov/genbank/statistics.
[accessed on Oct 21, 2014]
Guo, X., Meng, Y., Yu, N., & Pan, Y. (2014) ‘Cloud computing for Detecting High-Order
Genome-Wide Epistatic Interaction via Dynamic Clustering’ BMC Bioinformatics. 15 (102)
pp. 1471-2105.
Gurtowski, J., Schatz, M., & Langmead, B. (2013) http://www.bio-cloud-1449786154.us-
east-l.elb_amazonaws.com/cgi-bin/crossbow.pl/ [accessed 8 November 2014]
Hide, W. (2012) ‘The Promise of Big Data’. HSPH News.
http://www.hsph.harvard.edu/news/magazine/spr12-big-data-tb-health-costs. [accessed 31
August 2014].
Implementa. (2014) http://www.implementa.com/solutions [accessed 10 June 2014]
Lin, Y., Yu, C., & Lin, Y. (2013) ‘Enabling Large-Scale Biomedical Analysis in the Cloud’
BioMed Research International. Volume 2013 pp. 1-6.
Monica, N. & Kumar, R. (2013) ‘Survey on Big Data by Coordinating MapReduce to
Integrate Variety of Data’ International Journal of Science & Research . 2 (11). pp. 236-242.
Movirtu. (2014a) Movirtu Virtual SIM Solutions.
http://www.movirtu.com/#!manyme/c1aq3 [accessed 10 September 2014]
Movirtu. (2014b) Press Release. http://www.movirtu.com/#!090614-airtel-
movirtuagreement/c140h [accessed 10 September 2014]
Russell, J. (2012) ‘Hadoop’s Rise in Life Sciences’ Bio-IT World. pp. 1-7.
Schatz, M. C. (2009) ÇloudBurst: Highly Sensitive Read Mapping with MapReduce’
Bioinformatics. 25 (11) pp. 1363-1369.
Stanoevska-Slabeva, K. & Wozniak, T. (2010) ‘Çloud Basics- An Introduction to Cloud
Computing’ In: Dai, L., Xiao, J., Zhang, Z., Gao, X. & Guo, Y. (2012) ‘Bioinformatics
Clouds for Big Data Manipulation’ Biology Direct. 7 (43) pp. 1-7.
Telco Review. (2014) Will the virtual sim break the telco strangle
hold?http://techday.com/telco-review/news/will-the-virtual-sim-break-the-telco-strangle-
hold/195914/ [accessed 28 October 2014]

pp 217-230
Biographies
Cephas Mawere: Attained his BTech degree in Biotechnology at CUT, Zimbabwe in 2010.
Completed his MTech in Bioinformatics at JNTUH, India in September 2014. He is a HIT
Lecturer and Researcher. His research interests encompass the fields of Big Data, Cloud
Computing, Information Security and Computer-Aided Drug Design
Kudakwashe Zvarevashe: Attained his BSc degree in Information Systems at MSU,
Zimbabwe in 2010. Completed his MTech in Information Technology at JNTUH, India in
September 2014. He is a HIT Lecturer and Researcher. His research interests are in the areas
of Big Data, Information security, Cloud Computing and Web Services.

pp 217-230

MawereC- Ubuntunet paper publication 2015

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to MawereC- Ubuntunet paper publication 2015 (20)

MawereC- Ubuntunet paper publication 2015