bmc_article.tex

%% BioMed_Central_Tex_Template_v1.05
%%                                      %
%  bmc_article.tex            ver: 1.05 % %


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%                                     %%
%%  LaTeX template for BioMed Central  %%
%%     journal article submissions     %%
%%                                     %%
%%         <27 January 2006>           %%
%%                                     %%
%%                                     %%
%% Uses:                               %%
%% cite.sty, url.sty, bmc_article.cls  %%
%% ifthen.sty. multicol.sty		       %%
%%									   %%
%%                                     %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%                                                                 %%	
%% For instructions on how to fill out this Tex template           %%
%% document please refer to Readme.pdf and the instructions for    %%
%% authors page on the biomed central website                      %%
%% http://www.biomedcentral.com/info/authors/                      %%
%%                                                                 %%
%% Please do not use \input{...} to include other tex files.       %%
%% Submit your LaTeX manuscript as one .tex document.              %%
%%                                                                 %%
%% All additional figures and files should be attached             %%
%% separately and not embedded in the \TeX\ document itself.       %%
%%                                                                 %%
%% BioMed Central currently use the MikTex distribution of         %%
%% TeX for Windows) of TeX and LaTeX.  This is available from      %%
%% http://www.miktex.org                                           %%
%%                                                                 %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


\NeedsTeXFormat{LaTeX2e}[1995/12/01] \documentclass[10pt]{bmc_article}    



% Load packages
\usepackage{cite} % Make references as [1-4], not [1,2,3,4] \usepackage{url}  % Formatting web addresses
\usepackage{ifthen}  % Conditional \usepackage{multicol}   %Columns \usepackage[utf8]{inputenc} %unicode support
%\usepackage[applemac]{inputenc} %applemac support if unicode package fails \usepackage[latin1]{inputenc}
%%UNIX support if unicode package fails \urlstyle{rm}


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%	
%%                                             %%
%%  If you wish to display your graphics for   %%
%%  your own use using includegraphic or       %%
%%  includegraphics, then comment out the      %%
%%  following two lines of code.               %%   
%%  NB: These line *must* be included when     %%
%%  submitting to BMC.                         %% 
%%  All figure files must be submitted as      %%
%%  separate graphics through the BMC          %%
%%  submission process, not included in the    %% 
%%  submitted article.                         %% 
%%                                             %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%                     


\def\includegraphic{} \def\includegraphics{}



\setlength{\topmargin}{0.0cm} \setlength{\textheight}{21.5cm} \setlength{\oddsidemargin}{0cm}
\setlength{\textwidth}{16.5cm} \setlength{\columnsep}{0.6cm}

\newboolean{publ}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%                                              %%
%% You may change the following style settings  %%
%% Should you wish to format your article       %%
%% in a publication style for printing out and  %%
%% sharing with colleagues, but ensure that     %%
%% before submitting to BMC that the style is   %%
%% returned to the Review style setting.        %%
%%                                              %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


%Review style settings
\newenvironment{bmcformat}{\begin{raggedright}\baselineskip20pt\sloppy\setboolean{publ}{false}}{\end{raggedright}\baselineskip20pt\sloppy}

%Publication style settings \newenvironment{bmcformat}{\fussy\setboolean{publ}{true}}{\fussy}



% Begin ...
\begin{document} \begin{bmcformat}


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%                                          %%
%% Enter the title of your article here     %%
%%                                          %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\title{Six Questions and Answers Defining Cloud Computing for Digital, Sequencing-Based Biological Research}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%                                          %%
%% Enter the authors here                   %%
%%                                          %%
%% Ensure \and is entered between all but   %%
%% the last two authors. This will be       %%
%% replaced by a comma in the final article %%
%%                                          %%
%% Ensure there are no trailing spaces at   %% 
%% the ends of the lines                    %%     	
%%                                          %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


\author{Konstantinos Krampis\correspondingauthor$^{1}$% 
             \email{Konstantinos Krampis\correspondingauthor - agbiotec@gmail.com}%}
             \and
             Granger Sutton$^{1}$% 
             \and
             Vivek Sarangi
             }


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%                                          %%
%% Enter the authors' addresses here        %%
%%                                          %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\address{ \iid(1) Informatics Department, J. Craig Venter Institute, 9704 Medical Center Dr., Rockville, MD 20850, USA }

\maketitle

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%                                          %%
%% The Abstract begins here                 %%
%%                                          %%
%% The Section headings here are those for  %%
%% a Research article submitted to a        %%
%% BMC-Series journal.                      %%  
%%                                          %%
%% If your article is not of this type,     %%
%% then refer to the Instructions for       %%
%% authors on http://www.biomedcentral.com  %%
%% and change the section headings          %%
%% accordingly.                             %%   
%%                                          %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


\begin{abstract}
% Do not use inserted blank lines (ie \\) until main body of text.
\paragraph*{Background:} Text for this section of the abstract. 

\paragraph*{Results:} Text for this section of the abstract \ldots

\paragraph*{Conclusions:} Text for this section of the abstract \ldots \end{abstract}



\ifthenelse{\boolean{publ}}{\begin{multicols}{2}}{}

%%%%%%%%%%%%%%%%
%% Background %%
%%
\section*{Background}
\subsection*{Next-Generation Sequencing, Computing, and Digital Biological Research} 

Advances in recent years in the areas of high-throughput sequencing and synthetic genomics has defined "The 
Digital Age of Biology" \cite{venterdublin}, where Schrodinger's vision of "Life is code" \cite{Schrodinger1992} is 
now implemented in technologies that can be used to convert digital code into DNA that runs a living organism 
\cite{Gibson2010}. \pb

The latest generation of sequencing technologies is also being used in the area of metagenomics, where large-scale
studies of uncultivated microbial communities are performed. The J.  Craig Venter Institute (JCVI) for example has been
involved in several such metagenomic projects, including the Sorcerer II Global Ocean Sampling (GOS,
\cite{Rusch2007}) expedition to study marine microbial diversity, and also the National Institutes of Health
funded Human Microbiome Project to study human associated microbial communities \cite{Nelson2010}. \pb

Sequencing technologies continue to move in a direction where throughput per run is increasing while cost
per basepair is decreasing (review in \cite{Mason2012}). For instance, one of the most widely used instruments 
in the field currently, Illumina's GAIIx system can produce up to 95 Giga-base (Gb) of sequence per run \cite{Illumina} 
while the ABI SOLiD sequencer has yields of a similar range \cite{solid5500}. With the latest generation of 
instruments such as for example the HiSeq system, yield has reached 600 Gb \cite{Illumina}, while the Pacific 
BioSciences sequencer can produce 90 Gb in short amounts of time \cite{PacBio}.  \pb

Small-factor, benchtop sequencers are also available which can be acquired at a fraction of the cost, making them 
affordable for independent researchers running small laboratories. Examples in this category
include the GS Junior by 454, MiSeq by Illumina and Ion Proton by Life Technologies, providing sequencing 
capacity at 0.035Gb, 1Gb and 1.5Gb respectively for GS Junior, Ion Proton and MiSeq (review in \cite{Loman2012a}). 
That level of throughput is adequate for sequencing bacterial, small fungal or viral genomes and along with the
low cost per run (US \$225 -\$1100), sequencing has started to become a standard technique even in small laboratories.  
Example applications of sequencing for basic biological research include Single Nucleotide Polymorphism 
(SNP) variation discovery , gene expression analysis (RNAseq) and DNA-protein interaction analysis
(ChiPseq), (review in \cite{Mardis2008}). \pb

While sequencers generate datasets of significant size, they are typically bundled with only minimal computational and 
storage capacity for data capture during a run of the instrument.  For example, the un-assembled reads
returned from a single lane, single run of the Illumina GAIIx instrument after base calling are approximately 100 GigaByte
(GB) in size. Therefore, for laboratories acquiring a sequencer, scientific value cannot be obtained 
from this investment, unless it is accompanied by an almost equal or greater expense for informatics hardware 
infrastructure. In addition, significant software engineering bioinformatic data analysis expertise is required 
\cite{gogol2012overview}, in order to go from sequence data to valuable information such as assembled and 
annotated genomes. This means that besides large capacity computing servers, trained personnel competent to install, 
configure and use specific software to analyze and store  the generated data is required. Furthermore, public databases 
and software tools currently available online are not an option for downstream sequence data analysis, since 
BLAST \cite{altschul1990basic} available from NCBI for example \cite{johnson2008ncbi}, cannot accept input 
sequence files of more  than 0.5 GB  size for sequence similarity search.  \pb

An additional conundrum specific to bioinformatic analysis of sequencing data, is that computationally 
intensive tasks  that require extensive compute resources such as genome assembly or whole genome alignments
for example, are followed by genome annotation that is much less computationally demanding. This leads to sub-optimal utilization 
of computer hardware installed within data centers, while maintenance costs including electricity, cooling and salaries
for informatics support personnel are at constant or even increasing levels. For smaller academic institutions this can pose a significant
impediment in leveraging sequencing technology for research, as in addition to securing the funds for purchasing computing
hardware with adequate capacity to handle large-scale genomic datasets, they also need to maintain informatics systems
that are not utilized at its full capacity most of the time.  A second complication is related to the fact that 
databases with reference genomes \cite{Pruitt2009} are constantly growing in size, and for most bioinformatic analysis 
tasks they need  to be downloaded to local storage systems when performing comparative genome annotations. As these 
databases grow larger the process becomes more time consuming, incurring in higher bandwidth and storage costs for replicating 
the data locally.  Finally, building a data analysis infrastructure for next-generation sequencing also involves hiring trained 
bioinformatics engineers competent to implement specialized software tools and data analysis pipelines, which can incur 
higher costs than that of acquiring the computer hardware or maintaining a data center. \pb



%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Results and Discussion %%
%%

\section*{Results and Discussion}

\subsection*{What Role can Cloud Computing play for Digital Biological Research ?}

Computational data analysis can become a major bottleneck for smaller laboratories transitioning their experimental 
techniques to sequencing-based methods. Furthermore, sequencing instrument capacities follow an uptrend that surpasses 
that of Moore's Law \cite{schaller1997moore} while the cost per base pair follows an inverted trend, resulting in the genomics 
community to constantly take on sequencing projects of increasing scale and scope. Currently, most federally-funded projects 
conclude with upload of the sequences or annotated genomes to NCBI \cite{ncbi} databases such as the Sequence  
Read Archive (SRA,\cite{sra}). This provides little value  to researchers that do not have access to computational resources 
or informatics expertise, since multiple steps including data download, provision of high-performance computer servers and large-scale data storage, in addition to 
compiling and installing specialized bioinformatics software are required in order to utilize these datasets. \pb

As an alternative to investing in informatics infrastructure, researchers can rent computational 
and storage capacity from Cloud services such as Amazon EC2 \cite{awsec2}. This can potentially be
a better economic model for smaller research laboratories, as the cost for hardware and data center 
maintenance cannot be justified for only a few sequencing experiments. The Amazon EC2 Cloud,  
employs a charge model similar to traditional utilities such as electricity and users are billed based 
on the amount of computational capacity consumed on an hourly basis \cite{ec2price}. This particular Cloud 
service consists of thousands of computer servers with petabytes of storage, leveraging economies of 
scale to achieve low operational costs that in turn offers as savings to users. Furthermore, the Amazon Cloud 
has data centers in US East and West regions, European Union, Asia and Australia \cite{ec2regions}, providing researchers 
worldwide with the ability to tap into a large pool of computational resources, outside of institutional, 
economic or geographic boundaries. Overall, renting computational capacity from the Cloud has the 
potential to eliminate many of the upfront capital expenses for building information technology infrastructures 
for next generation sequencing, and result in transformation of the analysis and data processing tasks into well 
defined operational costs.  \pb

The 1000 Human genomes project \cite{clarke2012} has demonstrated a new approach for distributing genomic datasets 
in combination with computational capacity, by depositing sequence reads and reference genome mapping data on the 
Amazon Cloud storage. These datasets can be directly accessed by renting servers loaded with pre-installed bioinformatics 
software on the Cloud, and such examples have been provided by the Cloud BioLinux project \cite{1000tube1},  \cite{1000tube2}, 
\cite{Krampis2012}. Overall, by placing data from public projects on Cloud compute platforms, value for the community is 
immediately increased; no infrastructure other than a computer with internet access is required for a researcher to upload 
data generated at his or her laboratory , rent computational capacity and perform comparative analysis with 
data available on the Cloud storage. \pb

On the other hand, large research institutes with in-house bioinformatic core facilities that complete sequencing of new 
genomes on a regular basis, also have the resources and expertise to deliver more to the community than just sequence 
reads and annotated genomes uploaded to publicly-accessible repositories \cite{Pruitt2009}.  Currently, while many of the centers 
deposit software on open-source code repositories such as SourceForge \cite{sourceforge} or GitHub \cite{github}, an 
impediment for researchers trying to utilize the released software is the requirement to download, compile and configure all the 
dependencies including the type of operating system, software libraries, and computer hardware. This creates a bottleneck 
for laboratories lacking the required informatics expertise or computational infrastructure, that could be alleviated if bioinformatic
core facilities capture and distribute the bioinformatics expertise developed during each sequencing project on Cloud servers 
containing pre-configured bioinformatics tools, made publicly available along with the data on the Cloud. This approach,
can democratize access to computational resources required for high throughput sequencing data analysis and 
allow further  adoption of sequencing technology for basic biology research. Specialized compute servers with bioinformatics tools 
and data pipelines available on the Cloud through the Amazon EC2 service for example, could provide publicly accessible, 
high performance data analysis platforms for use by research groups acquiring sequencing capability. These platforms 
could be rented  on-demand at low cost, removing the need for implementing informatics infrastructure at each laboratory. \pb


\subsection*{Are Cloud-Based Bioinformatics Software Suites Available on the Cloud ?}
  
A number of systems that allow making bioinformatics tools accessible online through web portals have been 
developed during the past years, including the Biology Workbench \cite{Subramaniam1998Biology}, 
PISE \cite{Letondal2001Web}, wEMBOSS \cite{Sarachu2005WEMBOSS}, Mobyle \cite{Neron2009Mobyle}
BioManager \cite{Cattley2007BioManager} and BioExtract \cite{Lushbough2008Implementing}. Some 
of those systems are not actively developed anymore; most require significant software development effort 
and back-end system customization in order to deploy the tools through a web interface; with the exception of Mobyle, 
none of these portals provides users with a straightforward option to create and edit complex data analysis workflows; 
users have the option to download the source code for each system, but need to provision the hardware and
software engineering expertise to set up their own instance of the portal; large-scale data management or collaborative
data sharing and exchange among users is cumbersome, and available storage space is fized; finally, these portals do 
not leverage the Cloud's  scalability but are rather limited by the capacity of the server where the portal is installed, 
posing a limitation on the dataset size that can be processed.  \pb

On the other hand, centralized web portals for large-scale data analysis of bioinformatic datasets have been developed by 
well-funded institutions, including IMG/M \cite{Grigoriev2012}, CAMERA \cite{Altintas2010}, EBI \cite{Hunter2011} 
and MG-RAST \cite{Aziz2010}. While these portals are backed by considerable compute resources and data storage,  
their centralized nature is what eventually becomes a bottleneck. First, for those of the centralized portals that provide 
access to their resources, researchers  are required to apply for an account, in order to receive allocation of computational 
capacity for performing data analysis. Second, given the constantly increasing scale of genomic datasets, despite their 
capacity these portal cannot possibly support all small laboratories that purchase low-cost, benchtop sequencers and 
generate sequence data. Finally, a major drawback is that the software for most of these sites is not  open-source, while 
researchers often have to perform multiple submissions of their datasets, since each site offers a different sequence data 
analysis pipeline.  \pb 

An alternative to these centralized services are Cloud-based, scalable bioinformatics data analysis systems such as Galaxy
\cite{Goecks2010}, CloVR \cite{Angiuoli2011}, Cloud BioLinux \cite{Krampis2012} and BioKepler \cite{Altintas2011}. 
These systems are open-source and accessible to any laboratory or research group through Amazon EC2, but also are 
available for download an execution on private compute Clouds \cite{awsec2}. As an example, the Galaxy bioinformatics 
workbench includes a range of tools, from simple scripts that extract entries from sequence files, to complex algorithms 
for processing next-generation sequence data. Furthermore, Galaxy is a complete platform including a web portal software 
stack that provides the user interface for  executing the bioinformatics tools, in addition to an intuitive, drag and drop
canvas for composing  workflows and data analysis pipelines with the available tools. Finally, it provides a standardized 
method for easy deployment on the portal of command-line only bioinformatics software, by editing simple configuration 
files to specify the interface design \cite{galaxywiki}.  Through the Galaxy-Cloudman \cite{Afgan2010} 
framework, compute clusters for parallel data processing on Cloud services such as Amazon EC2 and private cloud can
be instantiated. \pb

Another community-centered, public access offering for computing on the Cloud is through our own work on Cloud Biolinux 
\cite{Krampis2012}, \cite{cloudbio}. This offering provides on-demand bioinformatics computing and a set of pre-configured 
sequence analysis tools within a high-performance Virtual Machine (VM) server that runs on a host of Cloud and virtualization 
platforms. The project is targeted to researchers that do not have access to large-scale informatics infrastructures for 
sequencing data analysis, but can instead rent on demand computational capacity from the Cloud. Users can access 
the tools by starting the Cloud BioLinux VM through the Amazon console web page \cite{console}, and  easily perform 
large-scale data analysis as we have demonstrated for examplewith the 1000 Human genomes \cite{1000tube1}, 
\cite{1000tube2},\cite{Clarke2012}.  Furthermore, the Cloud BioLinux VM is open-source, can be downloaded and modified, 
while advanced users can install  and run it on a private instance of the Eucalyptus \cite{euca} or Openstack \cite{openstack} Cloud platforms. A diverse 
community of researchers from both the US (Massachusetts General Hospital, Harvard School of Public Health, Emory 
University) and Europe (National Environmental Research Center, King's College London, Denmark Technical University, 
Netherlands Wageninen University) has been already established around the project \cite{googlegroup}. Finally, we have 
recently  expanded Cloud BioLinux by adding support for software developers through a framework for building 
and distributing bioinformatics VMs, that essentially provides a toolkit for implementation of customized, Cloud-based 
bioinformatics data analysis solutions. The framework includes a software management system that automates 
building a VM with a set of bioinformatics tools specified by the user and seamlessly deploys it across different
Cloud platforms, and is freely available from the GitHub code repository \cite{fabric}.  The overall goal is
to offer a platform for maintaining a range of specialized VM setups for serving different computing needs
within the bioinformatics community, and allow researchers to focus on the next challenges of providing data,
documentation, and the development of scalable analysis pipelines. \pb

Some more tool suites on the Cloud that have become available recently, including the non-profit/ open-source GenoSpace 
by Broad Institute \cite{genomespace} and SeqWare \cite{d2010seqware}, in addition to commercial offerings such as Illumina's 
BaseSpace \cite{basespace},  DNAnexus \cite{dnanexus} and Nimbus Informatics \cite{nimbusit}. GenomeSpace essentially integrates 
a set of tools and databases developed at the Broad Institute through  a unified graphical interface for end-users, in addition to 
offering Application Programming Interfaces (API) for programmatic access by developers, backed by Cloud VMs and storage.  Users 
can access the public GenomeSpace instance or create their local instance by retrieving the source code, in addition to accessing 
Virtual Machines (VM) with the complete system pre-installed and ready to execute with minimal configuration on the Amazon EC2 
cloud \cite{Afgan2010}. \pb SequeWare.... Similarly to GenomeSpace SeqWare provides source code and an Amazon Cloud VM 
with everything pre-installed and ready to execute with only minimal configuration. DNANexus currently includes tools for ChiPseq, 
RNAseq, 3'-end sequencing for expression quantification (3SEQ) and enzyme restriction analysis. DNAnexus runs on the Amazon 
Elastic Compute Cloud (EC2, online ref. 4), which provides on-demand virtual servers with various compute capacities.  \pb

The solutions presented above provide public access to scalable sequence data analysis resources for the genomic 
community, through which users can get access to pre-configured software and on-demand computing  using 
Cloud infrastructures. Nonetheless, specialized, high-performance bioinformatics applications and data pipelines 
implemented by bioinformatics core teams at large institutions, are usually coupled with specific hardware and 
the informatics infrastructure at each institution. As a result, significant effort might be required to refactor 
data analysis pipelines to run at a different site from where they were originally developed or port them on the cloud 
\cite{Wilkening2009}. Finally, while the current Cloud-based solutions are great for smaller laboratories that lack 
informatics resources and in addition the VM servers provide enhanced portability across sites, they are  simply a 
sophisticated container for bioinformatics software that in most cases has a monolithic design and does not leverage 
the distributed computing characteristics of the Cloud.  \pb


\subsection*{Unique characteristics of Cloud Computing versus traditional Bioinformatics Infrastructures}

One of the building blocks of cloud computing technology is virtualization \cite{Uhlig2005}, that allows entire 
compute servers including the operating system and all the necessary software packages for data analysis 
to be encapsulated within a Virtual Machine (VM). A VM is an emulation of a compute server, with virtual 
processors, memory and storage capacity, in the form of a single binary file that executes 
independently of the underlying hardware architecture, on both Cloud and desktop computers. 
Cloud services such as Amazon EC2 \cite{awsec2} provide high-performance computer hardware with a 
virtualization layer, on top of which users run VM servers. Since all software components and dependencies 
are  encapsulated  within the VM, it is possible to  distribute data analysis pipelines, databases, website portals, 
and all their required code libraries and configuration files in a ready to execute, compact and easy to 
download format.  This approach can remove many of the technical roadblocks encountered when 
performing  complex installations of open-source bioinformatics software, and consecutively make bioinformatics 
tools more accessible to the research community. \pb

In our experience with development of bioinformatics projects, it is difficult to provide long-term software
support or maintain web portals that provide online access to data analysis tools and databases, especially 
for projects funded by government grants that have an expiration date. Alternatively, by using Cloud VM servers
to build and maintain a bioinformatics system and subsequently create Whole System Snapshots (WSSE,
\cite{Dudley2010, Krampis2012}) of the VM servers, bioinformatics web portals and online databases that 
are build on the Cloud can be  preserved in their precise state when the snapshot was created. A snapshot 
essentially is a compressed,  exact replica of a VM server, capturing all of the software configuration, bioinformatics
pipelines, input data and sequence assemblies, genome annotations and all other sequence data analysis results. 
A snapshot is an executable binary file as the original VM, and by using it as a template the 
virtualization layer of a Cloud platform can instantiate multiple replicas of the original VM server \cite{ebs}. 
Finally, a researcher can set her snapshots to be publicly  accessible or share them only with specific users 
within the same Cloud, therefore providing access for collaborators to both data and software in a ready to execute 
and compact format. \pb

Regarding costs, the Amazon EC2 Cloud \cite{awsec2} charges for VM snapshots \$0.01 US  per GigaByte (GB) 
of storage used per month. Such low costs can allow researchers to allocate a relatively small amount compared 
to their overall informatics budgets for sequencing data analysis, and maintain a VM server snapshot for a number 
of years past the end of a funding cycle. For projects involving data release online though a web portal that is 
usually decommissioned shortly after funding for a project ends, using VM snapshots for archiving the portal on 
a Cloud platform, enables other researchers to lease compute time on the Cloud, and create fully-functional 
instances of the original VM server from the snapshots. Therefore, using VM technology and the Cloud for building 
bioinformatics systems and then creating compressed snapshots to reduce costs for long-term storage, offers 
an economical and flexible solution throughout and past the life cycle of a research project. \pb

Finally, use of virtualization and VM technology can provide two additional advantages: first, by
depositing data and pre-configured software on a publicly accessible, Cloud-based VM, allows for 
reproducibility, provenance and openness of the bioinformatics research. For example, following publication 
of assembly and annotation results from a genome sequencing project, researchers in the community 
might require to re-run the assembly with additional data generated at their own laboratory or to change algorithmic 
parameters and fine tune outputs such as gene predictions. Furthermore, lowering the barrier to access high-
performance informatics infrastructures required for working with next-generation sequencing datasets, is key 
for allowing researchers in the community to extract value from  data released from publicly funded projects, while 
also to  add value as similar studies take place. Second, by using the Cloud researchers have the capability to scale 
computational resources on-demand according to the amount of data generated from a sequencing project, 
through provision of the appropriate number of VM servers. With this approach usage of resources can be 
adjusted accordingly during the different analysis phases: while initially extensive computing resources will be 
necessary to perform assembly and annotation of the sequence data, computational resources (defined by
the number of running VM servers), can be scaled down for less computationally demanding tasks such as 
visualization and browsing of the sequence annotations. After funding has ended, a lab can further  lower the 
usage of the Cloud's computational resources and cut its informatics costs by archiving the VMs using snapshots 
\cite{Dudley2010}.  \pb


\subsection*{What the Public, Private, Open-Source or Commercial Cloud Solutions Available to Biologists Today ?} 

\subsubsection*{Accessing Computational Cycles on The Cloud}

Amazon Web Services (AWS, \cite{aws}) is one of the better established Cloud computing vendors, running on a
similar infrastructure with the one that powers Amazon.com's e-commerce web portals. Cloud computing services 
offered by this vendor that are most applicable to bioinformatics, include the Amazon Elastic Compute Cloud (EC2, \cite{awsec2}),  
Elastic Block Store (EBS, \cite{ebs}) and Simple Storage Service (S3, \cite{s3}), while a complete list of the available 
services can be found at \cite{aws}. These option provide respectively compute cycles through Virtual Machine (VM) 
servers, multiple virtual hard drives up to 1TeraByte (TB) that can attached to a running VM,  and web-accessible 
data storage. None of these options is tied to any specific operating system or programming model and each comes  
with affordable pricing, as for example a large capacity VM server with 64GB memory and 8 processor (CPU) cores 
that would suffice for many different types of bioinformatic analysis costs, \$2 US to rent per hour (for a complete price
 list see \cite{ec2price}). A large software developer community with a lot of expertise has formed during the past few 
 years around the Amazon Web Services discussion forums \cite{awsforums}, and based on our experience questions 
 regarding the  different services or requests for technical advice, are always answered in a day or less. \pb

Using the Cloud for bioinformatic data analysis is not limited only to the Amazon EC2 platform, since researchers
with access to a local computing cluster at their home institution have the option to run VM servers (and VM
snapshots downloaded from Amazon) on a private Cloud, such as Eucalyptus \cite{euca} or OpenStack \cite{openstack}. 
While OpenStack is the official Cloud of the Ubuntu Linux operating system \cite{ubuntucloud} and is included 
by default on a compute cluster that runs this particular Linux flavor, it can also be installed on clusters running other 
Linux versions \cite{openstackother}, and similarly for Eucalyptus \cite{eucalyptusother}. These Cloud platforms 
are essentially open-source replicas of Amazon EC2 and offer identical Application Programming Interfaces (API), 
meaning that applications developed on Eucalyptus or OpenStack will work seamlessly on the Amazon Cloud and vice-versa. 
This allows to seamlessly transfer VM server snapshots such as for example the Cloud BioLinux VM \cite{Krampis2012}  
across installations of these Cloud platforms, providing researchers with ready-to-execute  bioinformatics tools and data 
analysis pipelines pre-configured and installed on the VM. \pb

While Amazon Web Services \cite{aws} was the first vendor to offer public Cloud computing access at a large scale, 
many alternative platforms became available during recent years including the Google App Engine \cite{appengine}, Microsoft 
Azure \cite{azure}, GoGrid \cite{gogrid},  FlexiScale \cite{flexiscale} and the IBM SmartCloud \cite{smartcloud}.Multiple 
technical differences exist between these offerings, but they can be categorized into two broad groups: the first includes
Cloud platforms that provide users with access to Virtual Machines (VM) that are no different than standard Unix servers, available 
by Amazon, GoGrid, FlexiScale and IBM. On the other hand, Clouds that users access as abstracted computational resources that 
run software, without need for logging into Unix filesystems or provisioning VMs, are found on the Google and Microsoft offerings. 
This approach might sound as a better solution for smaller research groups with limited software engineering and informatics 
expertise, but is not without drawbacks. Specifically, the seamless execution of code and automatic scalability stems from the
fact that the Google and Microsoft Clouds require software developers to implement their applications using programming 
frameworks based on Python \cite{python} and .NET \cite{net} respectively. While this might be an option for the development of 
new software or web-based applications that are most suitable for these frameworks, the majority of existing 
bioinformatics software is designed to run on standard Unix servers and filesystems. In this case, Clouds that provide direct 
access to VMs are a better choice, while additionally a framework-specific Cloud implementation despite its advantages could 
eventually lead to vendor lock-in. \pb

Overall, Cloud platforms that provide access to VM servers can be a better choice for deploying existing software, in order 
to make it accessible to groups that do not have local informatics infrastructure but could instead rent compute time from 
the Cloud. In addition, for groups that have a longer-term vision for building Cloud-based bioinformatics infrastructures 
outside of traditional in-house data centers, portability across Cloud platforms should be a top criterion. While standalone 
software  can runs on VMs that can be easily converted and ported across public, private Clouds that provide a virtualization 
layer \cite{Krampis2012}, Cloud informatics infrastructures are essentially encoded in scripts that setup VM-based compute 
clusters and storage by issuing directives to the Cloud's API (\cite{Afgan2010}, \cite{scriptaws}). Therefore, for achieving 
portability of the complete infrastructure across different Cloud platforms with minimal software re-engineering effort, 
compatible and inter-operable APIs across the platforms is a requirement. Currently, this is only fulfilled amongst the open-source 
Eucalyptus / OpenStack Clouds, and the commercial Amazon Cloud. This fact should be carefully considered before choosing 
a Cloud provider for longer-term, especially if instances of the infrastructure is might be required to exist on both a private 
and a publicly-accessible, commercial Cloud. \pb

For researchers who do not have access to a compute cluster, neither have the available funds to lease computing time from 
Amazon, the government-funded Magellan Cloud \cite{magellan} provides an OpenStack cluster where researchers can apply 
for a user account. In addition, a number of academic computing centers in both the US and other countries have similar clusters 
with OpenStack installed \cite{openstackinstalls}, where scientistics could access a Cloud platform. Finally, an option for users is 
to to run the VM servers on a desktop  computer, using virtualization software such as VirtualBox \cite{vbox}, that is also 
open-source and can be installed in a single step on Windows, Mac or Linux  computers. The Cloud BioLinux project for example, 
provides  VM server snapshots that run  on both private clouds and VirtualBox \cite{cblorg}. \pb

Finally, professors that teach courses on Cloud computing, can apply for the Amazon Web Services educational grants 
\cite{awsgrants}, that provide free computing and data storage resources for educational use. Furthermore, the Amazon 
Cloud has established a program \cite{awspublicdata} that hosts free of charge a variety of large-scale public datasets that 
have significant value for the scientific community. For example, genomic datasets available through this program include 
the 1000 human genomes  data (\cite{Clarke2012}, \cite{aws1000gen}) the NCBI flu genomes \cite{awsflu} and Ensembl 
human genome annotation database \cite{awsensembl}. Users of the Amazon Cloud can access, copy, and perform computation 
on the  data using VM servers, and just pay for the compute and storage resources they lease. \pb

\subsubsection*{Accessing Storage Capacity on The Cloud}

Cloud data storage services provide the advantage of data centers distributed across the globe, and therefore can make available 
connection endpoints in various geographic regions that reduce network transfer latency for data exchange among remotely 
located research sites. The Amazon S3 storage for example \cite{s3} has data centers located on the US East and West coast, 
European Union (Ireland), South America (Brazil), and Asia-Pacific (Japan, Singapore and Australia). A researcher working with
large-scale sequencing datasets can choose to upload data to their closest data center, then initiate replication across the 
different regions through the S3 CloudFront service \cite{cloudfront}, and subsequently allow collaborators worldwide 
to retrieve the data from the nearest location. Furthermore,  data stored using this service are protected from physical disasters,
since they replicated  by default across three different regions. Note that triple-replication is not synonymous with backup that
prevents accidental deletion by  the user, but instead refers to protection from permanent loss of a single data center.  The S3 
storage model deviates from that of POSIX-compliant hard drives \cite{mathur2007new}, but instead uses data objects that have unique Uniform Resource 
Location (URL)  identifiers across the Amazon Cloud and are also web browser-accessible. In detail, S3 is organized using a two-level 
namespace with top-level folders called "buckets", and while each Amazon account may have up to 100 buckets, a bucket can store 
an unlimited number of data objects. A file on S3 with open access permissions can be accessed by simply pasting its URL on a web 
browser, but for upload, modification, bulk operations or for files that require authentication, one of the two Application Programming 
Interfaces (API) must be used to access the service: the first is based on the Representational State Transfer protocol (REST, 
\cite{fielding2000})  and the second on the Simple  Object Access Protocol (SOAP, \cite{soap}), both with a rich set of programmatic
access code libraries available for developers \cite{s3developer}. Alternatively, users can interact with the service through various 
desktop client applications that have graphical front-ends, available for all operating systems (\cite{cyberduck}, \cite{s3browse}, 
\cite{s3fox}). \pb

A second storage option on the Amazon Cloud is Elastic Block Store (EBS,\cite{ebs}), that provides POSIX-compliant \cite{mathur2007new} 
hard drives up to 1 TeraByte (TB), multiples of which can be attached to a running VM. While these data volumes persist after a Virtual 
Machine (VM) shutdown and can be re-attached to a new VM booted at a later time, data stored on EBS are not triple-replicated and are 
stored only on a single Amazon data center. This makes them prone to loss during physical disasters or hardware failures. Another 
option for data storage is the transient virtual hard-drive available as the file system of a running VM, but given that a VM would be 
used by a cost-conscious researcher only during execution of software (unless it is a 24/7 uptime web server), its storage should 
only be used as a temporary holding for the software's data outputs.  \pb

Overall, storage costs can be reduced by maintaining on high-availability, low-latency storage such as Amazon S3 only non-processed 
data for which value will be generated following compute. Alternatively, for data that have been already processed and is loss is not 
critical an option is the S3 Reduced Redundancy Storage \cite{reduced}, that allows users to cut down on the costs  by storing 
non-critical, reproducible data at lower levels of redundancy and disaster-protection than Amazon S3’s standard storage. Finally, if 
data preservation is important but the data are rarely accessed, the Amazon Glacier service \cite{glacier} provides an archival storage
option at one-tenth of the cost. \pb

\subsection*{Which Factors Challenge Adoption of Cloud-Based Solutions for Bioinformatics }

In the case of Cloud computing and similarly to a large-scale institutional server clusters, besides availability of 
compute resources that in both of these cases should not be a concern, three major facts affect users planning 
to access  the computational system for data analysis: first, flexibility of moving data inputs and outputs to and 
from the system; second, available interfaces for the user interacting with the system, and third; since both the 
Cloud and institutional clusters are multi-tenant systems, what are the mechanisms safeguarding eacy user's data
integrity, privacy and isolation. While many studies are available in the literature that review available large-scale 
institutional clusters based on these criteria (ref,ref,ref), here we attempt to provide an overview of the Cloud
in these respect, using as basis the Amazon EC2 platform.
 
Another important concern for Cloud-based bioinformatic tools is related to the data transfer bottleneck from
the local sequencing machines to the Cloud servers.  According to the data published for the Amazon Cloud
platform (online ref.11), 600GB of data would require approximately one week to upload on to the remote Cloud
servers, when using an average broadband connection of 10Mbps. With a faster T3 connection which is usually
easily obtainable even at small research institutions, within one week 2TB of data can be uploaded or
approximately 600GB in 2 days.  Solutions addressing this issue are available both as software that maximizes
data transfer over the network compared to traditional File Transfer Protocol (FTP), or physical disk drive
import/export services offered by the Cloud provider to its customers. Aspera's server (online ref. 12) has
been recently integrated to NCBI's infrastructure, and researchers can download a free client that allows
increased upload speeds to the Short Read Archive (online ref. 13).  Through the Aspera software, transfer
bandwidth between NCBI and the European Bioinformatics Institute for data sharing in the 1000 Genomes Project,
has been increased from 20Mbps to 1000Mbps (see online ref. 14). 

Finally, the Amazon offers the option for its users to physically ship disk drives to the company's offices
and have the data copied to their servers (online ref. 11). With only 80 import cost for disk drives up to 4TB
of data (4000GB), this is the most efficient method if we take into account the charge by Amazon for 0.10 per
GB of bandwith consumed, which would add up to 60 for a 600GB data upload. In addition to that cost, the
expense for obtaining a high-bandwidth internet connection for the data upload should be taken into account.
We expect the Microsoft Azure Cloud platform to offer a similar service in the near future, given the requests
on the Azure developer forums and the immediate consideration of the matter by Microsoft (online ref. 15).

For researchers that would like to leverage the advantages of a VM with the pre-installed assembly portal for
working with the completed assemblies but consider the public Cloud as not secure option, we will offer the
alternative of returning to them by mail an external hard drive with a VM containing with the portal and
assembly data. Users will then be able to load and execute the VM on a local computer cluster with a
Eucalyptus/OpenStack Cloud or on a PC using Virtualbox. We are currently offering a similar solution with
Cloud BioLinux [3], where the project's VM is available for download and execution on a local Cloud or a PC
from our website [4].  Upon local execution of a VM users will simply need to point their browser to the
portal's Internet or local IP address [55] assigned automatically by either the Cloud or Virtualbox (Fig.1B).
The IP address is available through each Cloud platform's or the Virtualbox software administrative
interface, and we will provide extensive documentation (see subsequent paragraph) on uploading, running and
accessing a local VM on the different platforms by extending the available Cloud BioLinux project
documentation.

An intuitive \cite{youtube}
our VM with the pre-installed tools on the Amazon EC2 Cloud, they will only need to follow four simple steps
through their web browser: visit the Amazon Cloud website and create a new account, start the VM execution
wizard through the Cloud's control console [58], choose computational capacity for the VM (memory,
processor, cores, storage capacity), and specify username and password credentials for accessing the running
VM. Each running VM receives a unique web address, and by using their web browser to access the address, a
researcher can login to the portal interface with the assembly tools. These four steps are described in detail
in our Cloud BioLinux publication and the project's documentation [59].

In the Cloud BioLinux work we combined the convenience of SaaS for end-users with the power of cloud computing, 
in order to bring pre-installed specialized bioinformatics application which need large computational capacity 
such as those for genome assembly while simplifying the the way users can get onto the cloud.
A user can start and access the OSMF Frame VM instance in three simple steps by using the Amazon EC2 cloud console 
graphical user interface that is accessible via a web browser: first the user signs up for an Amazon EC2 account and 
after she obtains the credentials logins to the cloud console (http://aws.amazon.com/console);  within the Amazon 
console the users clicks the “Launch Instance Wizard” button and specifies the OSMF Frame VM volume identifier 
(our project website will provide the VM identifier for the most recent update, but the latest VM will be also identifiable 
by the meta-data added to the volume); following the steps of the wizard within the web browser the users selects 
computational capacity and storage for the OSMF VM, and specifies a username and password for the OSMF 
WebInterface login (Fig.1, additional users can be created after the initial login); finally, once the wizard steps are 
complete and the VM status shows “running”, the user copies the assigned URL address of the VM from the Amazon 
cloud console in a new web browser window in order to access the OSMF interface. Through the URL users can get access
to CloudMan and Galaxy (ref Enis and Brad)The process of starting a VM on 
the cloud and connecting to it has been documented for the JCVI Cloud BioLinux VM instances (REF), but nonetheless 
more detailed documentation, video tutorials and user support will be available from the proposed project's website 
and discussion forum (see Education & Outreach section).

\subsection*{Bioinformatics Computing and Science as A Service on the Cloud ?}

Science as a Service (ScaaS) for bioinformatics, can be defined along similar lines with the Software as a Service  
computing model (SaaS, \cite{papazoglou2003}), where software is running on remote datacenters. In this model, 
users access the software through a web browser or a desktop client application and there is no requirement other 
than a desktop computer with Internet connection, since the SaaS service provisions and manages the computing 
infrastructure, in addition to setting up the software and all its dependencies. \pb

A Science as a Service (ScaaS) model could be of benefit to the bioinformatics community, where currently a unified approach is
not available for researchers to access software or datasets, most of which have been generated as a result of federal grants 
awarded to individual investigators \cite{Stein2010}.  The different approaches range from websites created
by small laboratories and which provide online access to specialized bioinformatic tools,  to centralized web portals 
developed by large institutions such as NCBI \cite{ncbi} where web-based versions of mainstream applications including BLAST 
\cite{altchul}  are available. Nonetheless, in most cases datasets and software as source code are simply made available for 
download from FTP sites. In every case there are significant drawbacks including restrictions on the size of datasets that can be 
processed on the websites provided by small laboratories, as they are backed by compute servers with limited computational 
capacity; on the other hand, web portals set up by large institutions are also restricted on the computational resources that 
they can offer to the public (for example, users cannot upload multiple Gbp of sequence to the NCBI-BLAST website), while also
NCBI cannot possibly provide online access to all available bioinformatic tools; finally, in the case of software available only as 
source code, provisioning informatics infrastructure and the technical expertise for performing specialized installation 
procedures can be a burden for non computationally-savvy investigators, with a prime example being genome assembly software.  \pb


Edit section on additional ScaaS (tools suites on Cloud), that has been moved on the tool suites on Cloud Section.

For users of the Cloud who require more control and additional flexibility to customize the computational infrastructure 
were their software is running, an alternative computing model is Infrastructure as a Service (IaaS, \cite{bhardwaj2010cloud}).
The Amazon Elastic Compute Cloud (EC2) is one of the most popular providers with this model, and essentially has become 
a standard for IaaS service providers followed on the open-source replicas OpenStack and Eucalyptus (\cite{openstack}, 
\cite{euca}). The Amazon Cloud uses Virtual Machine (VM) servers as the basic unit for computational resource allocation, 
that are available in different capacities \cite{instancetypes} allowing users the option to lease a portions of the underlying 
physical compute server capacity according to their budget and data processing needs. The specific term for the Amazon VMs is 
Amazons Machine Images (AMIs), and while these  run on top of a virtualization layer user interaction is no different than 
accessing a physical server with a full operating system, processors and memory depending on the VM capacity selected. In
a few recent studies VM performance characteristics such as read-write speed of the virtual hard drives, processors speeds and 
network latency for inter-communication of nodes of cluster instantiated using EC2 VMs, was found to be lagging that of  
in-house built compute clusters and specialized network interconnects such as Myrinet or Infiniband (\cite{jackson2010}, 
\cite{hill2009quantitative}, \cite{boden1995myrinet}, \cite{infiniband2000infiniband}). Nonetheless, users can rent at higher 
cost specialized VMs \cite{instancetypes} that are connected within the Cloud through high-speed network or connect to 
virtual hard  drives that are physically backed by Solid State Disks (SSD), and which have shown promising results \cite{jackson2010}. 
Overall, other than specialized scientific applications that require specialized networking and configuration within a cluster, 
the on-demand availability of the Amazon EC2 Cloud service can provide a viable alternative to dedicated clusters, as it is 
similar to commodity hardware clusters often built in smaller labs without the expense or labor the build process requires. \pb






%%%%%%%%%%%%%%%%%%%%%%
\section*{Conclusions} Text for this section \ldots




%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section*{Authors contributions} Text for this section \ldots



%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section*{Acknowledgements} \ifthenelse{\boolean{publ}}{\small}{} Text for this section \ldots



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%                  The Bibliography                       %%
%%                                                         %%              
%%  Bmc_article.bst  will be used to                       %%
%%  create a .BBL file for submission, which includes      %%
%%  XML structured for BMC.                                %%
%%                                                         %%
%%                                                         %%
%%  Note that the displayed Bibliography will not          %% 
%%  necessarily be rendered by Latex exactly as specified  %%
%%  in the online Instructions for Authors.                %% 
%%                                                         %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


{\ifthenelse{\boolean{publ}}{\footnotesize}{\small} \bibliographystyle{bmc_article}  % Style BST file
\bibliography{bmc_article} }     % Bibliography file (usually '*.bib') 

%%%%%%%%%%%

\ifthenelse{\boolean{publ}}{\end{multicols}}{}


\end{bmcformat} \end{document}