Accelerating Genomic Data Generation and Facilitating Genomic Data Access Using Decentralization, Privacy-Preserving Technologies and Equitable ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Page 1 of 23 Accelerating Genomic Data Generation and Facilitating Genomic Data Access Using Decentralization, Privacy-Preserving Technologies and Equitable Compensation Dennis Grishin,1,2,3 Kamal Obbad,1 Preston Estep,1,3 Kevin Quinn,1 Sarah Wait Zaranek,3 Alexander Wait Zaranek,1,3 Ward Vandewege,1,3 Tom Clegg,3 Nico César,3 Mirza Cifric,1,3 George Church1,2,3 Authors 1 Nebula Genomics, Inc., San Francisco, USA; 2Department of Genetics, Harvard Medical School, Boston, USA; 3 Veritas Genetics, Inc, Danvers, USA. Corresponding Author Dennis Grishin, Nebula Genomics Inc., 73 Sumner Street, #401, San Francisco, CA 94103, USA; dgrishin@g. harvard.edu Keywords: Arvados, Blockchain, Data Privacy, Data Sharing, DNA Sequencing, Genomics, Homomorphic Encryption, Nebula Category: Use Cases/Pilots/Methodologies In the years since the first human genome and technical challenges impede the growth of was sequenced at a cost of over $3 billion, genomic data and hinder data sharing. technological advancements have driven the price below $1,000, making personal In this article, we propose that these challenges genome sequencing affordable to many can be addressed by combining decentralized people. Personal genome sequencing has the system design, privacy-preserving technologies, potential to enable better disease prevention, and an equitable compensation model in a more accurate diagnoses, and personalized platform that vests control over data with therapies. Furthermore, sharing genomic data individual owners; ensures transparency and with researchers promises identification of the privacy; facilitates regulatory compliance; causes of many diseases and the development minimizes expensive data transfers; and shifts of new therapies. However, sequencing costs, the sequencing costs from consumers, patients, data privacy concerns, regulatory restrictions, and biobanks to researchers in industry and Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 2 of 23
Figure 1—Human genome sequencing cost, 2001–2017.
academia. We exemplify this by describing the using DNA hybridization microarrays. These
implementation of Nebula, a distributed genomic tests are referred to as genotyping and they
data generation, sharing, and analysis platform. assess the presence or absence of genetic
variants associated with certain traits. For
The Human Genome Project has sequenced a cost less than $100, genotyping typically
and assembled the first human reference reads out only ~0.02% of the human genome,
genome at a cost of over $3 billion.1 Since then, at predefined positions, often missing
development of next-generation sequencing health-relevant genetic variants that must be
technology has resulted in exponentially reported. In addition, variant identification
decreasing sequencing cost (Figure 1).2 Today, at a small number of positions does not
the sequencing of a whole human genome allow discovery of novel variants, including
costs less than $1,000. This price is projected those that cause disease; the majority of
to drop to $100 in the next few years.3 The these variants are distributed throughout
exponentially decreasing DNA sequencing the genome and remain undiscovered.4 This
costs have made personal genome sequencing limits the usefulness of genotyping data to
affordable to patients as well as healthy researchers.
individuals.
OPPORTUNITIES
Personal genome sequencing is becoming As genomic sequencing becomes more affordable,
more common as prices decline, but most it opens up opportunities for individuals as well as
genetic tests to date have been performed researchers in academia and industry.
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 3 of 23
Personal genome sequencing can support personal genomics company 23andMe received
data-driven decision-making for health-related $60 million from Genentech16 and $300 million
issues. Studies estimated that ~2% of people from GlaxoSmithKline17 for access to genotyping
carry genetic variants that cause or predispose data collected from its customers. Other
them to a wide variety of diseases at various biopharma companies have launched their own
levels of severity, the majority of which can be sequencing projects. AstraZeneca announced it
preventable or treatable.5 In addition, every parent would sequence 2 million human genomes,18 and
carries, on average, approximately five genetic Regeneron is leading a $100 million consortium
variants that might cause diseases in offspring to sequence approximately 500,000 samples
if the other parent carries the same variant.6 The collected by the UK Biobank.19
presence of certain genetic variants also has
been associated with adverse effects for ~7% of CHALLENGES
Food and Drug Administration-approved drugs.5 Multiple obstacles hinder the realization of
Personal genome sequencing can also help opportunities offered by personal genomics.
healthy individuals make better lifestyle choices. Many people are deterred by the costs of
For example, genetic variants have been shown personal genome sequencing, as well as concerns
to cause sensitivities to certain nutrients7–9 and over genomic data privacy. Research is hampered
to increase risks of sports-related injuries.10–12 In by the resultant scarcity of genomic data and is
the future, advancement in understanding human further compounded by difficulties with respect
genetics will make personal genome sequencing to data access.
more insightful, while correcting pathological
genetic variants will become possible as more In 2018, the number of genotyped people
and more gene therapies enter clinical trials.13 surpassed 10 million and is expected to grow to
more than 100 million by 2021.20 This growth
Researchers study genomic data sets to is driven by a combination of factors, notably
identify genetic variants that cause diseases. consumer interest in ancestry analysis coupled
This enables the research and development with a decrease in genotyping costs below
of therapies targeting disease-associated $100.21 In contrast, consumer interest in whole
genes with increasing specificity. Genomics- genome sequencing has grown slowly due to
guided therapeutic discovery has been applied a significantly higher cost. A recent survey
successfully to many types of cancers, rare revealed that only ~3% of people are willing to
genetic diseases, and, increasingly, common pay >$1,000 for whole genome sequencing.22
complex diseases.14 Furthermore, genomics- For the majority of consumers, whose primary
guided patient cohort recruiting can reduce interest in the area can best be described as
the failure rate of clinical trials by enriching nonmedical “infotainment,” the benefits of
for likely responders and reducing reducing sequencing over genotyping do not justify the
adverse reactions. This approach to clinical trials significantly higher cost.
promises to reduce surging drug development
costs and lead to more drugs reaching the market At the same time, the surge in popularity of
and benefiting patients.15 genetic testing, forensic utilization of genetic
databases,23 and the purchase of genetic data
These opportunities are recognized by the by biopharma companies24 have increased
biopharma industry. For example, the leading consumer and media attention to genetic
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 4 of 23
data privacy. Studies show that privacy concerns genomic data sharing, privacy-preserving
are legitimate, as data sharing policies of computing has been utilized to protect genomic
many personal genomics companies do not data privacy, and different compensation models
fulfill transparency guidelines with regard have been explored to incentivize genomic data
to the confidentiality or sharing of customer sharing.
genetic data.25 These developments are likely
to exacerbate reported privacy concerns over Genomic Data Sharing
genetic data26,27 and deter personal genomic The GA4GH Beacon Project33 and i2b2
sequencing. SHRINE34 are two of the most advanced
systems for biomedical data sharing. Both are
For researchers, low adoption of personal networks that enable participating institutions
genome sequencing has resulted in low to connect their genomic (and clinical) databases
availability of genomic data. According to and process queries about the presence of
estimates, only ~500 thousand human genomes genetic variants and traits, including medical
had been sequenced by 2017.3 This is detrimental conditions. This federated model minimizes
for research because very large genomic expensive data transfers and enables institutions
data sets are necessary to find links between to retain control of their data. This addresses
genetic variants and traits, such as disease privacy, regulatory, and technical challenges
predispositions. Finding such links is difficult that are associated with centralized storage and
because most traits are the product of complex transfers of genomic data.
interactions of many genetic variants, while the
effects of individual genetic variants are, on However, there are limitations. First,
average, very small.28 Low diversity of genomic functionality is currently limited to simple
data sets further compounds the search for links queries. Orchestrated, distributed computations
between genetics and disease.29 required for data processing and analysis are
currently not supported. Second, participation
The scarcity of genomic data is exacerbated by is limited to academic research institutions and
difficulty in data access due to fragmentation of hospitals. There are no patient- or consumer-
genomic data across proprietary data silos.30 Data focused portals that would enable individuals
sharing is further hindered by the large size of to easily contribute their personal genomic
genomic data, which impedes data transfer over data. Third, decentralized governance and
networks.31 In addition to logistic and technical compensation mechanisms have not been
challenges, data access is often complicated by implemented.
restrictive government regulations that hinder
data sharing.32 Low availability of genomic Genomic Data Protection
data combined with data silos also results in Distributed genomic data storage and computing
high prices, making it unaffordable to many can help protect genomic data privacy. However,
researchers. data owners cannot always maintain in-house
servers and therefore they often must outsource
PREVIOUS WORK data storage and computing to third parties,
Solutions to the challenges outlined above have such as cloud service providers. To protect
been proposed previously. Federated data storage the privacy of genomic data that are shared
systems have been implemented to facilitate with untrusted third parties, encryption-based
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 5 of 23
privacy-preserving techniques have been adopted to the challenges described in the previous
for genomics. These techniques enable third sections. This model requires consumers to pay
parties to execute computations and return results for genetic testing and result interpretation,
without having access to plaintext genomic data. while personal genomics companies often take
ownership of the generated genomic data and
Privacy-preserving techniques have been applied sell it to biopharma companies (Figure 2). This
previously to distributed medical and genomic model requires consumers to carry the costs
databases. For example, MedCo integrates and relinquish ownership and control of their
with the i2b2 SHRINE framework and uses a genomic data, which discourages genetic testing.
homomorphic data encryption scheme to enable In addition, this model promotes genomic data
outsourcing of genomic data storage and query fragmentation across private data silos, which
execution to untrusted third parties.35 Another hampers data access and increases data prices.
example is the Secure Multi Party Query
Language framework that implements similar We propose to combine and extend previous
functionality and privacy guarantees using secure work on genomic data sharing networks, privacy-
multiparty computations.36 Data can also be preserving technologies, and compensation
protected using trusted hardware. An example models to create a new model for personal
is the PRINCESS framework that executes genomics that may overcome these challenges
computations on genomic data inside protected (Figure 3).
memory regions of Intel microprocessors.37
First, the functionality of genomic data sharing
Compensation Models networks must be extended beyond simple
Over the past few years, personal genomics queries. This requires a network that can be
companies have explored different models to integrated with a full-fledged bioinformatics
compensate individuals for contributing their platform that supports genomic data processing
personal genomic data to research studies. In and analysis. Implementing this functionality
2016, Genos offered to help its customers sell would bundle fragmented genomic data and
their genomic data to researchers.38 A similar make it available for analysis on a single
model that uses a cryptocurrency instead of fiat network, thereby facilitating data access for
money was adapted by EncrypGen in 2017.39 researchers.
Most recently, LunaDNA announced that it
would compensate genomic data contributors Second, the data sharing network must expand
with company stock.40 These models are similar beyond research institutions and must be
in that individuals who want to participate must accessible to individuals who want to share their
already own their personal genomic data, or personal genomic data. However, the resulting
choose to purchase genetic testing because of the network decentralization will necessitate a more
prospect they will be rewarded later for sharing democratic governance model. This potentially
the data. can be achieved by integrating blockchain
technology, which holds the promise of enabling
PERSONAL GENOMICS 2.0 decentralized, self-governing networks.
The traditional model for genomic data
generation and sharing that has been adopted by Third, the privacy of genomic data must be
most personal genomics companies contributes protected. Data access control on the blockchain
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 6 of 23
Figure 2. The traditional model for genomic data generation and sharing.
can ensure transparent consent management, also may result in a reduction in genomic data
while privacy-preserving technologies can help prices and thus empower more researchers to
protect shared genomic data. Together with access large genomic data sets.
the distributed computing model that “brings
algorithms to the data,” these technologies can DESIGN CONSIDERATIONS
enable network participants to retain ownership To implement a system as outlined in
and control of their genomic data, thereby the previous section, one must integrate
reducing privacy concerns and incentivizing data a bioinformatics platform that supports
sharing. distributed data storage and computing with a
suitable blockchain framework, as well as with
Fourth, genome sequencing and data sharing techniques for privacy-preserving computing.
also must be incentivized by implementing Here, we review and evaluate existing options.
subsidy and compensation mechanisms. The
decentralized data sharing model can facilitate Bioinformatics Platforms
this, as it enables researchers to connect directly Bioinformatics platforms have been developed to
with individuals with traits of interest, subsidize facilitate organization of genomic data; to enable
their genome sequencing costs, and compensate parallelized, high-performance computing with
them for data sharing. Elimination of middlemen support for complex dependencies; and to allow
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 7 of 23 Figure 3—Alternative model for personal genomics that may overcome challenges. a modular pipeline design that is flexible and containerization of computing environments, ensures reproducible results.41 Table 1 shows a and utilization of standardized application comparison of popular bioinformatics platforms. programming interfaces (APIs). The development of bioinformatics platforms Based on these considerations, Arvados and has been driven by exponentially growing DNAstack appear as suitable choices for the genomic data and marked by adaption of proposed genomic data sharing platform. Both multiple computing trends. Storage and platforms have an API-focused architecture and processing of genomic data has moved from data sharing functionality. DNAstack integrates local servers to remote clouds. This has enabled with the GA4GH Beacon Network, while scalable data storage and computing and Arvados supports platform-agnostic, federated facilitated access sharing to genomic data sets. cloud environments and has an open-source To scale beyond single clouds, efforts are being codebase. made to create federated cloud environments that could enable distributed data storage Blockchain Frameworks and computing.48,49 Furthermore, the growth Blockchain technology has three use cases of genomic data and development of new in the proposed system. First, the need to bioinformatics tools that must be integrated provide transparent consent management can into workflows are driving the development of be addressed by the ability of blockchains standardized workflow description languages, to store data access permissions on an Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 8 of 23
Table 1. Comparison of bioinformatics platforms
Criteria Arvados42,43 DNAstack44 Seven Bridges45 DNAnexus46 Galaxy47
Hardware Federated Google Cloud with Clouds Clouds Local servers
clouds and Beacon Network
servers integration
Pipeline design API-based; API-based; web Web GUI Web GUI Web GUI
web GUI GUI
Containers Yes Yes Yes Yes Yes
Workflow CWL WDL CWL Custom Custom
language
Open source Yes No No No Yes
Platform launch 2013 2014 2012 2010 2005
year
API: application programming interface; CWL: Common Workflow Language; GUI: graphical (rather than textual) user interface;
WDL: Workflow Description Language.
immutable public ledger. Second, blockchains Based on these requirements, permissioned
can enable implementation of decentralized blockchains frameworks such as Exonum and
systems governed by network participants. Hyperledger Fabric appear most suitable (Table 2).
Third, an immutable ledger can facilitate Hyperledger Fabric has been more widely adopted,
verification of the integrity of decentrally but Exonum offers transparency and security that
stored data. is comparable to public blockchains.
Based on these use cases, one can create a First, Exonum-based blockchains offer public
set of requirements that a suitable blockchain read access but restrict write access to selected
framework must fulfill. First, consent validator nodes. By making read access to the
management requires that the identity of blockchain public, transaction audit does not
researchers who request to access data are rely on trusted parties. Exonum transactions
known to data owners. To this end, network are verified in real time by all nodes. Thus,
access must be limited to data buyers whose all network participants are able to audit the
identity has been verified. Therefore, consent blockchain state collectively.
management requires a blockchain that supports
permissioned access. Second, Exonum supports anchoring of
transaction logs in the Bitcoin blockchain.
Second, a large, decentralized data marketplace Hashes of the Exonum blockchain state are
requires smart contract functionality and high periodically written to the Bitcoin blockchain, so
transaction throughput. Private blockchains can even if all permissioned Exonum nodes collude,
achieve higher transaction throughputs than the transaction history cannot be falsified unless
public blockchains because the ability to write the attacker succeeds in compromising the
transactions to the blockchain is limited to a Bitcoin blockchain as well.
group of permissioned validator nodes. However,
this makes private blockchains more centralized Third, Exonum uses a byzantine fault-tolerant
and less dependable. (BFT) consensus algorithm that protects against
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 9 of 23
Table 2. Comparison of blockchain frameworks
Criteria Exonum50 Hyperledger Fabric51 Ethereum52
Read access Public Private Public
Write access Private Private Public
Consensus Byzantine fault-tolerant Fault-tolerant (FT) Proof of work (PoW)
(BFT)
Transactions per ~3,000 ~3,000 ~15
second (TPS)
Smart contracts Yes (Rust, Java) Yes (Go, Java) Yes (Solidity)
Light clients Yes No Yes
Public blockchain Yes No NA
anchoring
Open source Yes Yes Yes
NA: not applicable.
malicious behavior of permissioned nodes. amounts during the computation. It is possible,
In contrast, Hyperledger and other private however, to improve the performance of fully
blockchains rely on less computationally homomorphic encryption and secure multiparty
intensive fault-tolerant (FT) consensus computations significantly if they are optimized
algorithms that protect against node breakdown for specific use cases. Practical performance
but not malicious behavior. Exonum offers both levels have been demonstrated for queries on
BFT consensus and high transaction throughput genomic data54,55 and genome-wide association
because it is written in Rust, one of the fastest studies (GWAS).56
programming languages. Furthermore, Rust
offers memory safety which eliminates many Alternative technologies have drawbacks of
vulnerabilities that are commonly exploited by their own. Intel Software Guard Extensions
hackers. technology is a hardware-assisted approach
that protects data privacy by executing
Privacy-Preserving Technologies computations inside private memory regions. It
Table 3 shows a comparison of privacy- offers good performance but has been affected
preserving technologies that all have been by vulnerabilities that can compromise data
applied to secure genomic data.53 Fully privacy.57 Differential privacy methods protect
homomorphic encryption and secure multiparty data privacy by introducing randomness.
computations enable computations on encrypted However, obfuscation of computation results can
data that generate encrypted results. These complicate interpretation of studies.53
encrypted results, when decrypted, correspond to
the results of the same computation on plaintext NEBULA
data. However, fully homomorphic encryption In this section, concepts and design
is very slow and typically suffers from very considerations outlined in the previous sections
large ciphertext expansion. The limitation of are illustrated by describing the technical
secure multiparty computation protocols is implementation of Nebula—a decentralized
that they require transfers of very large data genomic data generation, sharing, and analysis
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 10 of 23
Table 3. Comparison of privacy-preserving technologies
Fully Secure Intel Differential
Homomorphic Multiparty Software Guard Privacy
Criteria Encryption Computations Extensions
Principle Computations Distributed Computations inside Introduction of
(additions AND computations on private memory randomness to
multiplications) ciphertexts regions data/results of
on ciphertexts computations
Computation Very slow Slow Fast Fast
time
Memory usage Very high High Low Very low
Communication High Very high Low Low
cost
Specific None None Vulnerabilities have Noise makes
limitations been discovered; interpretation
requires Intel CPUs of results more
difficult
CPU: central processing unit.
platform. Nebula integrates the Arvados42,43 intention is to preserve data privacy by enabling
bioinformatics platform (github.com/curoverse/ investigators to query the whole database
arvados) with the Exonum50 blockchain and discover their data of interest, without
framework (github.com/exonum) and a fully compromising the privacy of the queried data.
homomorphic data encryption scheme (Figure 4). In the future, it should be possible to extend the
application of privacy-preserving technologies to
Arvados has two core services: Keep and GWAS and other computations.
Crunch. Keep is a distributed content-
addressable storage system that enables scalable The Nebula blockchain is an Exonum-based
storage of genomic big data, high throughput blockchain through which the Nebula network
data access, and efficient data management. will be governed, consent will be documented,
Crunch is a workflow management engine and the data will be secured. Exonum-based
that enables flexible creation and parallelized blockchains have three types of nodes: auditors,
execution of data analysis pipelines and light clients, and validators. Auditors are
generation of reproducible results. Arvados full nodes that maintain a copy of the entire
implements a distributed data storage and blockchain content and can generate transactions.
computing model that minimizes required data Light clients also can generate transactions, but
transfers. This helps address big data challenges, they replicate only information that is relevant
regulatory restrictions, and data privacy risks. to them instead of the whole blockchain content.
Validators are permissioned nodes that verify
Utilization of a homomorphic data encryption transactions received from auditors and light
scheme enables implementation of privacy- clients and write new blocks to the blockchain.
preserving queries on genomic data. The While the current implementation of Nebula
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 11 of 23
Figure 4—Overview of the Nebula platform.
uses the Exonum framework, other permissioned payments to their wallets by operating light
blockchains, in particular Hyperledger Fabric, clients on the Nebula blockchain.
can be used as well.
• Network maintainers are organizations
that operate validator nodes on the Nebula
The Nebula network has four types of
blockchain. Validator nodes will collectively
participants: data owners, network maintainers,
control data access by managing encrypted
data buyers, and storage and compute providers.
key shares, verifying transactions, and keeping
track of data stored in Keep and computations
• Data owners can be private individuals or
executed by Crunch.
institutions. They will store encrypted genomic
data in public or private clouds that are part • Data buyers are researchers who wish to
of the Keep storage system. They will be able obtain access to genomic data. They will be
to control access to their data and receive operating auditor nodes to keep a local copy of
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 12 of 23
the metadata, which they will use to locate data To this end, the Nebula platform enables a
stored in Keep, verify data integrity, and keep data buyer to create a smart contract that
track of access permissions. Data buyers will specifies the blockchain addresses of data
be able to query homomorphically encrypted owners previously identified in a query and
data, utilize smart contracts to acquire data send cryptocurrency tokens to that smart
access permissions from data owners, and use contract. The data owners are notified that
Crunch to run analysis pipelines. a buyer has offered to pay their sequencing
costs. If a data owner accepts the offer by
• Storage and compute providers are data executing the smart contract, the deposited
owners that operate private clouds, or third tokens are sent to a sequencing provider.
parties that offer storage and computing Next, the data owner receives a saliva
services (e.g., Google, Amazon, and collection kit and submits a saliva sample
Microsoft). They will form a federated cloud to the sequencing facility. The sample
environment that hosts the Keep storage is sequenced, and the genomic data are
system and Crunch-managed containers deposited on a Keep server specified by
within which computations are executed. the data owner. Data hashes, along with
The development of Nebula is ongoing. Some blockchain addresses of all data owners and
parts of the platform, in particular, Arvados, buyers, are written to the blockchain. The
have been fully implemented over the past data buyer who paid the sequencing costs is
few years and are already being deployed by permitted to access and analyze the data. The
various organizations. Other parts of Nebula, in data owner receives interpretations of his
particular, the homomorphic encryption schemes, genomic data and is able to share data access
are a relatively recent addition and are not yet with additional data buyers.
fully integrated. A report on the progress of our
work was published in a white paper.58 Here we
describe the implementation of Nebula in greater Phenotypic data
detail but also revise some previously made Information about medical conditions and
design choices. other traits is referred to as phenotypic data.
These data are generated primarily through
Data Generation survey questions. The platform utilizes a
phenotyping toolkit that maps plain-language
Genomic data survey responses to clinical descriptions
Personal genome sequencing cost is a significant in Human Phenotype Ontology (HPO)59
factor in preventing more widespread consumer format. Survey data can be verified using two
adoption. Therefore, a key consideration in the approaches. First, comparing the incidence of
design of the Nebula platform was to provide a medical conditions in the general population to
mechanism to shift sequencing costs from data the incidence observed in the platform’s data
owners (e.g., consumers and biobanks) to data set will enable identification of survey results
buyers (e.g., pharma and biotech companies). that deviate from the expected results. Second,
This is being implemented by enabling data survey data can be verified by referencing
buyers to query the Nebula database, identify Electronic Health Records (EHRs) imported
data sets of interest, and pay the sequencing costs through the Fast Healthcare Interoperability
to generate and access genomic data (Figure 5). Resources (FHIR) API.
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 13 of 23 Figure 5—Genome sequencing subsidy payment on the blockchain. Data Encryption Encryption Standard (AES) encrypted. The AES Privacy of genomic and phenotypic data are keys are encrypted with validator public keys and protected through client-side encryption by bundled with the encrypted data. data owners and encryption key management by blockchain validator nodes. To enable data Data Storage buyers to discover data prior to purchasing data access, the platform implements a lattice-based Data fully homomorphic encryption scheme. To this Genomic data are stored in Keep, a distributed end, blockchain validator nodes generate public– content-addressable storage system that retrieves private key pairs and construct a single collective files based on their content. Addresses of files public key (Figure 4). Data owners encrypt their are generated through cryptographic digest of survey responses and genetic variant lists with their content. Keep combines content-addressing the collective public key and upload them to with the distributed storage architecture of the a Keep server. The homomorphic encryption Google File System.60 Keep splits encrypted files scheme protects data privacy by enabling data into 64-megabyte blocks and stores them in an buyers to execute Structured Query Language underlying object store or file system (Figure 6). (SQL)-like queries on the homomorphically The content addresses of the blocks are stored on encrypted data. Files that contain raw sequencing the blockchain and are used to find data locations data and are not used for queries are Advanced and check data integrity. Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 14 of 23
Figure 6—Data blocks are stored in Keep. Block hashes are stored on the blockchain.
Keep is designed for storing genomic and other Storj can potentially be supported if computing
types of biomedical big data. First, its content- on stored data becomes possible. Data owners
addressing offers high-speed storage and retrieval can register new, personal cloud instances or
by eliminating an indexing service, a potential store their encrypted data in shared clouds.
bottleneck and point of failure, and enabling Based on phenotypic information, data sets that
direct connections between the storage and are likely to be analyzed together are stored in
compute subsystems. Second, content-addressing physical proximity, which minimizes slow and
works well for data written to disk once and read expensive data transfers.
many times, a characteristic of genomic data,
as it does not change over time but is accessed As sequencing data are processed, different
frequently. Third, fixed-size data blocks allow file formats are generated and stored in Keep.
scalable distributed storage of big data, and Typically, Keep stores FASTQ files that contain
content-addressing enables easy file verification, raw sequencing data (~200 gigabytes/genome),
which is particularly important for distributed Binary Alignment Map files that store aligned
databases. sequencing reads (~100 gigabytes/genome),
and Variant Call Format files that store genetic
Keep is designed to be a distributed, hybrid variants (~200 megabytes/genome). Additionally,
storage system. Data owners can choose to Nebula uses the Compact Genome Format (CGF)
store their data in clouds such as Amazon Web to generate compact genomic data summaries.
Services (AWS), Google Cloud Platform (GCP), Genomes in the CGF format are represented by
and Microsoft Azure, or on private bare-metal pointers referencing sequences in a tile library
servers. Decentralized file storage solutions such (Figure 7). CGF offers a consistent, standardized
as InterPlanetary File System (IPFS), Sia, and representation of genomic data that makes
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 15 of 23
Figure 7—Simplified representation of a tile library and a Compact Genome Format (CGF) file. The
rectangles represent tile variants at different positions and the dotted line illustrates the tile composition of
specific genome.
different types of sequencing and genotyping transactions, add new blocks to the blockchain,
data interoperable. The CGF representation and update the key-value store. Storage of
is also very space efficient (~30 megabytes/ metadata on an immutable ledger helps secure
genome), which facilitates file transfers, and the integrity of the decentralized Nebula
enables fast queries and efficient analysis.60 database. To this end, multiple column families
are implemented:
Tabular phenotypic data generated through
surveys and imports of EHRs are stored in • Data ownership is registered by assigning each
physical proximity with associated genomic data. block content address the blockchain address of
In contrast to static genomic data, phenotypic the data owner who added the block to Keep.
information is much more dynamic and smaller • Data locations are described by assigning each
than genomic data. This makes utilization of the block content address the Uniform Resource
Google File System and content addressability Locator (URL) of a Keep server.
unsuitable. Therefore, phenotypic data files are • Data integrity is verified by re-hashing data
stored as Not only SQL documents. blocks and comparing their hashes with content
addresses that are stored on the blockchain.
• Data buyer identities, including names and
Metadata institutional affiliations, are verified, linked
To organize data stored in Keep, Nebula stores to blockchain addresses, and stored on the
metadata on the blockchain in a key-value store. blockchain.
When new data are added to Keep or existing • Access permissions to the Nebula platform and
data are modified, blockchain transactions data stored in Keep also are managed on the
are generated. Validator nodes verify these blockchain.
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 16 of 23
Data Discovery criteria, as well as their blockchain addresses.
Utilization of fully homomorphic encryption is This enables data buyers to connect with data
intended to address the privacy barrier to data owners to pay sequencing costs or to purchase
sharing. It enables data owners to make their data access to existing genomic data (Figure 8).
available for discovery without privacy risks,
while at the same time allowing data buyers Data Analysis
to explore the database before purchasing data The Arvados container and pipeline management
access to perform analyses. engine, Crunch, executes computations on data
stored in Keep. Crunch implements a distributed
To this end, data buyers will construct a SQL-like computing model whereby workflows, and not
query and encrypt it with the collective public the genomic data, are moved between cloud
key that has been constructed by validator nodes instances whenever possible. Highly distributed
and used to encrypt phenotypic information genomic data processing is possible because
and genetic variant lists. The encrypted query is many intensive bioinformatics computations,
executed on homomorphically encrypted data such as alignment and variant calling, are
and an encrypted result is generated. The query performed on single genomes and are easily
result is re-encrypted by the validator nodes parallelizable. To this end, Crunch executes
under a public key provided by the data buyer tasks inside Docker containers that are created
and shared with data buyer who can now decrypt physically close to the data locations in Keep
it with its private key. A query can return the and distributes computations between many
number of data owners that matched the specified processing units.
Figure 8—Secure data discovery through queries on homomorphically encrypted data.
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 17 of 23 Crunch ensures result reproducibility through Security standardization of computing workflows Homomorphic encryption can enable privacy- using Common Workflow Language (CWL),61 preserving queries for data discovery. However, which enables connection of open-source most computations that are necessary for and proprietary bioinformatics software into typical genomic data analysis workflows do not workflow pipelines that are flexible, portable, achieve practical runtimes when executed on and scalable. Crunch can access CWL pipelines homomorphically encrypted data. Therefore, stored in public or private Git repositories such other security mechanisms must be utilized. as GitHub. CWL can be used to implement end-to- Platform access control end bioinformatics analysis pipelines. To protect data owners and their data, data Typically, CWL pipelines include common, buyers are required to go through a partially computationally intensive “secondary analysis” decentralized, three-step permission process. tasks, such as alignment of sequencing reads to a The first step is data buyer authentication. reference genome and variant calling. However, Here, a blockchain validator node will verify a “tertiary analysis” tasks, which often involve data buyer’s personal and institutional identity. computing on genomic data sets rather than Blockchain addresses of verified data buyers will single genomes and are less standardized, also be added to the blockchain metadata store. Data can be incorporated into CWL pipelines. Typical buyers will then be able to connect to Nebula examples are statistical tests that are used in REST API servers and use Crunch to execute GWAS to identify correlations between genetic pipelines on data stored in Keep. Data buyer variants and phenotypes. For such tertiary authentication will enable data owners to verify analysis tasks, Nebula uses Lightning,62 a system data buyer identity before agreeing to share for high-performance, in-memory computations data access. Furthermore, immutable storage of on genomic data in the CGF. Lightning integrates data buyer identities on the blockchain enables into CWL pipelines and enables fast queries and identification of data buyers who have violated execution of complex machine learning tasks on consent agreements or have bypassed pipeline large genomic data sets. execution control. CWL pipelines can also be used to analyze and interpret personal genomic (and phenotypic) Pipeline execution control data. First, users can build their own custom To protect data privacy, the platform design pipelines to analyze their personal data and incorporates the ability to define approved also share pipelines among each other using bioinformatics tools and CWL workflows. public Git repositories. Second, developers The intent is to prevent data buyers from can build and monetize genomic apps. To this downloading genomic data or executing any end, CWL pipelines can be stored in private computations that attempt to extract a large repositories, and access by Crunch may require amount of information about individual data a smart contract-mediated token payment to the owners. This approach was chosen because it pipeline developer. The approach of bringing has the ability to provide a sufficient level of apps to the data facilitates protection of personal data privacy protection without significantly information. restricting data buyers. Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 18 of 23 Data access control re-encrypt the data under the data buyer’s public The first task in every CWL pipeline is to get key. Finally, the data buyer’s access permission access to the input data (Figure 9). Here, a is registered on the blockchain and tokens are data buyer executes a smart contract on the sent from the smart contract to the data owner’s blockchain. The inputs are the data buyer’s wallet. Crunch can now load decrypted data into blockchain address and the content addresses of a Docker container and begin pipeline execution. all data blocks of the input files. The data buyer also deposits tokens inside the smart contract Governance and defines a token payout for data access. When The Nebula blockchain can be used to enable a data owner’s light client synchronizes with Nebula network participants to collectively the blockchain, the data owner is notified of the govern the network, in particular, to help data access request. The data owner can decide maintain data protection. To this end, for about data sharing based on offered payment and example, Token-Curated Registries (TCRs)63 identity of the data buyer. The data owner grants can be used to conduct elections that determine data access by executing the smart contract. validator nodes or whitelist data analysis The blockchain validator nodes then verify the pipelines. TCRs are lists that are curated integrity of the requested data stored in Keep by decentrally by token holders. Importantly, comparing data hashes with the content addresses economic incentives drive the token holders to stored on the blockchain and collectively curate the list’s contents judiciously. In brief, Figure 9—Data access control and data purchases. Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34
Page 19 of 23
network participants can cast votes whereby the produced the data. If the source of the genomic
weights of votes scale with their token holdings. data is unknown, or the sequencing facility
Since token holders are invested in the network, does not cooperate, genomic data cannot be
they are incentivized to maintain its proper validated. A possible solution to this problem
operation that ensures data protection. can be a model that compensates personal
genomics companies and other genomic data
DISCUSSION producers for validating data authenticity. Data
The obstacles that hinder personal genome collected from different sources also must be
sequencing and genomic data sharing have a made interoperable. It is particularly challenging
significant impact on the progress of research to curate health records and other types of
into disease prevention, drug development, phenotypic data. However, standards such as
and other crucial aspects of human health. We FHIR are being developed very actively and
described one approach to overcoming these have already enabled applications that can collect
obstacles, using a combination of multiple EHRs across different health systems.65
technologies. A number of challenges remain
to be addressed regarding data privacy, data The idea of a personal data marketplace is very
validation, data curation, and data economics. new and has not yet been implemented at scale.
A personal data marketplace is likely to differ
Data privacy protection requires decentralization from traditional marketplaces in important ways.
of data generation and further development of For example, data supply can be regarded as
privacy-preserving technologies. Today, genomic being unlimited because an individual can share
data generation is limited to laboratories that data access with an unlimited number of data
own expensive sequencing machines operated buyers. Personal data marketplaces also would
by experienced technicians. Centralized genomic be asymmetric, since individuals are likely to be
data generation leads to data privacy risks that unaware of the value of their personal data and
may be averted if sequencing is decentralized. are thus at risk of not being compensated fairly.
We anticipate that this will become possible soon The novelty of personal genomic data further
as new technologies are being developed that compounds these challenges and makes market
would enable compact, affordable, and easy-to- dynamics difficult to predict. We anticipate
operate sequencing machines.64 Data privacy that future research into economics of data
protection is also impaired by current limitations marketplaces will help answer these and other
of privacy-preserving technologies that do not open questions.
allow complex computations on large data sets
and require extensive optimization for every Contributions: Dennis Grishin, Kamal Obbad,
use case, which hinders effective data analysis. and Kevin Quinn wrote the article. Dennis
However, practicality of privacy-preserving Grishin, Kamal Obbad, and George Church
technologies has been steadily increased over developed the ideas described in the article.
the past few years, and we anticipate continuing Dennis Grishin, Kamal Obbad, and Kevin
progress in the future. Quinn are leading the development of the
Nebula platform. Alexander Wait Zaranek, Ward
Data validation and curation are another area of Vandewege, Tom Clegg, Nico César, Preston
challenge. Validation of genomic data requires Estep, and Mirza Cifric contributed to the
assistance of the sequencing facilities that have development of Arvados. Alexander Wait Zaranek
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 20 of 23
and Sarah Wait Zaranek developed the Compact 7. Yang A, Palmer AA, de Wit H. Genetics
Genome Format and Lightning. All authors edited of caffeine consumption and responses
and/or reviewed the final article. to caffeine. Psychopharmacology. 2010
Aug;211(3):245–57.
Acknowledgments 8. Mattar R, de Campos Mazo DF, Carrilho
The authors would like to thank Armon Rahim FJ. Lactose intolerance: Diagnosis, genetic,
and clinical factors. Clin Exp Gastroenterol.
and Nathaniel Tucker for their help in reviewing
2012 Jul 5;5:113–21.
this article. 9. Freeman HJ. Risk factors in familial forms
of celiac disease. World J Gastroenterol.
Conflict of Interest 2010 Apr 21;16(15):1828–31.
All authors are employees, advisors, or 10. O’Connell K, Knight H, Ficek K, et al.
collaborators of Nebula Genomics. Interactions between collagen gene variants
and risk of anterior cruciate ligament
Funding Statement rupture. EJSS. 2015;15(4):341–50.
Development of the Nebula platform is funded 11. Tiziano FD, Palmieri V, Genuardi M,
by Nebula Foundation. Zeppilli P. The role of genetic testing in
the identification of young athletes with
inherited primitive cardiac disorders at risk
REFERENCES of exercise sudden death. Front Cardiovasc
1. International Human Genome Sequencing Med. 2016 Aug 26;3:28.
Consortium. Finishing the euchromatic 12. Bennett ER, Reuter-Rice K, Laskowitz
sequence of the human genome. Nature. DT. Genetic Influences in Traumatic Brain
2004 Oct 21;431(7011):931–45. Injury: Chapter XI. In: Laskowitz D, Grant
2. Wetterstrand KA. DNA Sequencing Costs: G, editors. Translational Research in
Data from the NHGRI Genome Sequencing Traumatic Brain Injury. Boca Raton, FL:
Program (GSP) [Internet]. [cited 2018 Jan CRC Press/Taylor and Francis Group; 2015.
11]. Available from: https://www.genome. 13. Ginn SL, Amaya AK, Alexander
gov/sequencingcostsdata/ IE, Edelstein M, Abedi MR. Gene
3. Illumina Promises to Sequence Human therapy clinical trials worldwide to
Genome For $100—But Not Quite Yet. 2017: An update. J Gene Med. 2018
2017. [cited 2018 Oct 10]. Available May;20(5):e3015.
from: https://www.forbes.com/sites/ 14. Cardon LR, Harris T. Precision medicine,
matthewherper/2017/01/09/illumina- genomics and drug discovery. Hum Mol
promises-to-sequence-human-genome-for- Genet. 2016 Oct 1;25(R2):R166–72.
100-but-not-quite-yet/#672a5d72386d 15. Rojahn SY. Genomics could blow up the
4. Maurano MT, Humbert R, Rynes E, et al. clinical trial. MIT Technology Review
Systematic localization of common disease- [Internet]. 2013 Nov 12 [cited 2018
associated variation in regulatory DNA. Aug 25]; Available from: https://www.
Science. 2012 Sep 7;337(6099):1190–5. technologyreview.com/s/521496/genomics-
5. Lindor NM, Thibodeau SN, Burke could-blow-up-the-clinical-trial/
W. Whole-genome sequencing in 16. Herper M. Surprise! With $60 Million
healthy people. Mayo Clin Proc. 2017 Genentech Deal, 23andMe Has A Business
Jan;92(1):159–72. Plan [Internet]. Forbes. 2015 [cited 2017
6. Berg JS, Adams M, Nassar N, et al. Oct 1]. Available from: https://www.forbes.
An informatics approach to analyzing com/sites/matthewherper/2015/01/06/
the incidentalome. Genet Med. 2013 surprise-with-60-million-genentech-deal-
Jan;15(1):36–44. 23andme-has-a-business-plan/
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 21 of 23
17. Bloomberg. GlaxoSmithKline Is Acquiring 26. Bloss CS, Ornowski L, Silver E, et al.
a $300 Million Stake in 23andMe [Internet]. Consumer perceptions of direct-
Fortune. [cited 2018 Aug 25]. Available to-consumer personalized genomic
from: http://fortune.com/2018/07/25/ risk assessments. Genet Med. 2010
glaxosmithkline-23andme-gsk/ Sep;12(9):556–66.
18. Ledford H. AstraZeneca launches project to 27. Sanderson SC, Brothers KB, Mercaldo ND,
sequence 2 million genomes. Nature. 2016 Clayton EW, Antommaria AHM, Aufox SA,
Apr 28;532(7600):427. et al. Public attitudes toward consent and
19. Herper M. Drug company consortium to data sharing in Biobank Research: A large
sequence the genes of 500,000 Britons over multi-site experimental survey in the US.
next two years. Forbes Magazine [Internet]. Am J Hum Genet. 2017 Mar 2;100(3):
2018 Jan 8 [cited 2018 May 27]; Available 414–27.
from: https://www.forbes.com/sites/ 28. Visscher PM, Wray NR, Zhang Q, et al.
matthewherper/2018/01/08/drug-company- 10 years of GWAS discovery: Biology,
consortium-to-sequence-the-genes-of- function, and translation. Am J Hum Genet.
500000-britons-over-next-two-years/ 2017 Jul 6;101(1):5–22.
20. Khan R, Mittelman D. Consumer genomics 29. Popejoy AB, Fullerton SM. Genomics
will change your life, whether you get is failing on diversity. Nature. 2016 Oct
tested or not. Genome Biol. 2018 Aug 13;538(7624):161–4.
20;19(1):120. 30. Lawler M, Maughan T. From Rosalind
21. Allyse MA, Robinson DH, Ferber MJ, Franklin to Barack Obama: Data sharing
Sharp RR. Direct-to-Consumer Testing 2.0: challenges and solutions in genomics and
Emerging models of direct-to-consumer personalised medicine. New Bioeth. 2017
genetic testing. Mayo Clin Proc. 2018 Apr;23(1):64–73.
Jan;93(1):113–20. 31. Feltus FA, Breen JR 3rd, Deng J, et al.
22. Marshall DA, Gonzalez JM, Johnson FR, The widening gulf between genomics data
et al. What are people willing to pay for generation and consumption: A practical
whole-genome sequencing information, and guide to big data transfer technology.
who decides what they receive? Genet Med. Bioinform Biol Insights. 2015 Sep
2016 Dec;18(12):1295–302. 23;9(Suppl 1):9–19.
23. Kolata G, Murphy H. The golden state 32. Majumder MA, Cook-Deegan R, McGuire
killer is tracked through a thicket of DNA, AL. Beyond our borders? Public resistance
and experts shudder. The New York Times to global genomic data sharing. PLoS Biol.
[Internet]. 2018 Apr 27 [cited 2018 Aug 2016 Nov;14(11):e2000206.
21]; Available from: https://www.nytimes. 33. Global Alliance for Genomics and Health.
com/2018/04/27/health/dna-privacy-golden- GENOMICS. A federated ecosystem for
state-killer-genealogy.html sharing genomic, clinical data. Science.
24. Ducharme J. A major drug company now 2016 Jun 10;352(6291):1278–80.
has access to 23andMe’s genetic data. 34. Weber GM, Murphy SN, McMurry AJ, et
Should you be concerned? Time [Internet]. al. The Shared Health Research Information
2018 Jul 26 [cited 2018 Aug 21]; Available Network (SHRINE): A prototype
from: http://time.com/5349896/23andme- federated query tool for clinical data
glaxo-smith-kline/ repositories. J Am Med Inform Assoc. 2009
25. Laestadius LI, Rich JR, Auer PL. All Sep;16(5):624–30.
your data (effectively) belong to us: Data 35. Raisaro JL, Troncoso-Pastoriza J,
practices among direct-to-consumer Misbach M, et al. MedCo: Enabling
genetic testing firms. Genet Med. 2017 secure and privacy-preserving exploration
May;19(5):513–20. of distributed clinical and genomic
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34Page 22 of 23
data. IEEE/ACM Trans Comput Biol 46. DNAnexus Documentation [Internet].
Bioinform [Internet]. 2018 Jul 13; [cited 2018 Oct 10]. Available from: wiki.
Available from: http://dx.doi.org/10.1109/ dnanexus.com
TCBB.2018.2854776 47. Goecks J, Nekrutenko A, Taylor J, Galaxy
36. Bater J, Elliott G, Eggen C, Goel S, Kho Team. Galaxy: A comprehensive approach
A, Rogers J. SMCQL: Secure querying for supporting accessible, reproducible,
for federated databases. Proceed VLDB and transparent computational research in
Endowment. 2017 Feb;10(6):673–84. the life sciences. Genome Biol. 2010 Aug
37. Chen F, Wang S, Jiang X, et al. PRINCESS: 25;11(8):R86.
Privacy-protecting rare disease International 48. Chaterji S, Koo J, Li N, Meyer F, Grama
Network Collaboration via encryption A, Bagchi S. Federation in genomics
through software guard extensions. pipelines: Techniques and challenges. Brief
Bioinformatics. 2017 Mar 15;33(6):871–8. Bioinform [Internet]. 2017 Aug 29; [cited
38. Molteni M, Allain R, Chen S, Thompson 2018 Oct 10]. Available from: http://dx.doi.
A, Simon M, Gonzalez R. Genos will org/10.1093/bib/bbx102
sequence your genes—And help you sell 49. Workflow Execution Service (WES) API
them to science. Wired [Internet]. 2016 [Internet]. Github; [cited 2018 Oct 11].
Dec 15 [cited 2018 Oct 6]; Available from: Available from: https://github.com/ga4gh/
https://www.wired.com/2016/12/genos-will- workflow-execution-service-schemas
sequence-genes-help-sell-science/ 50. Exonum Documentation [Internet]. [cited
39. Lin P. Blockchain: The missing link between 2018 Oct 10]. Available from: exonum.
genomics and privacy? Forbes [Internet]. com/doc
2017 May 8 [cited 2018 Oct 6]; Available 51. Androulaki E, Barger A, Bortnikov V, et al.
from: https://www.forbes.com/sites/ Hyperledger fabric: A distributed operating
patricklin/2017/05/08/blockchain-the-missing- system for permissioned blockchains. In:
link-between-genomics-and-privacy/ Proceedings of the Thirteenth EuroSys
40. Brown KV. Share your DNA, get shares: Conference. New York: ACM; 2018. pp.
Startup files an unusual offering. Bloomberg 30:1–30:15. (EuroSys ’18).
News [Internet]. 2018 Oct 5 [cited 2018 Oct 6]; 52. Wood G. Ethereum: A secure decentralised
Available from: https://www.bloomberg.com/ generalised transaction ledger. 2014.
news/articles/2018-10-05/illumina-backed- [Internet]. [cited 2018 Oct 10]. Available
startup-asks-sec-to-let-it-pay-people-for-dna from: https://ethereum.github.io/
41. Leipzig J. A review of bioinformatic yellowpaper/paper.pdf
pipeline frameworks. Brief Bioinform. 2017 53. Aziz MMA, Sadat MN, Alhadidi D, et al.
May 1;18(3):530–6. Privacy-preserving techniques of genomic
42. Arvados Documentation [Internet]. [cited data-a survey. Brief Bioinform [Internet].
2018 Oct 10]. Available from: doc.arvados.org 2017 Nov 7; [cited 2018 Oct 10].
43. Zaranek AW, Clegg T, Vandewege W, Available from: http://dx.doi.org/10.1093/
Church GM. Free factories: Unified bib/bbx139
infrastructure for data intensive web 54. Çetin GS, Chen H, Laine K, et al. Private
services. Proc USENIX Annu Tech Conf. queries on encrypted genomic data. BMC
2008 May 1;2008:391–404. Med Genomics. 2017 Jul 26;10(Suppl 2):45.
44. DNAstack Documentation [Internet]. [cited 55. Sousa JS, Lefebvre C, Huang Z, et al.
2018 Oct 9]. Available from: https://docs. Efficient and secure outsourcing of genomic
dnastack.com/java-sdk/ data storage. BMC Med Genomics. 2017 Jul
45. Seven Bridges Documentation [Internet]. 26;10(Suppl 2):46.
[cited 2018 Oct 10]. Available from: docs. 56. Cho H, Wu DJ, Berger B. Secure genome-
sevenbridges.com/docs wide association analysis using multiparty
Blockchain in Healthcare TodayTM ISSN 2573-8240 online https://doi.org/10.30953/bhty.v1.34You can also read
