A Codon Frequency Obfuscation Heuristic for Raw Genomic Data Privacy

A Codon Frequency Obfuscation Heuristic for Raw Genomic Data Privacy
Kato Mivule
Computer Science Department
Bowie State University
kmivule@gmail.com
Abstract
Genomic data provides clinical researchers with
vast opportunities to study various patient ailments.
Yet the same data contains revealing information,
some of which a patient might want to remain
concealed. The question then arises: how can an
entity transact in full DNA data while concealing
certain sensitive pieces of information in the genome
sequence, and maintain DNA data utility? As a
response to this question, we propose a codon
frequency obfuscation heuristic, in which a
redistribution of codon frequency values with highly
expressed genes is done in the same amino acid
group, generating an obfuscated DNA sequence. Our
preliminary results show that it might be possible to
publish an obfuscated DNA sequence with a desired
level of similarity (utility) to the original DNA
sequence.
Keywords: Genomics data privacy; DNA
obfuscation; Codon frequency table; Data privacy
modeling.
1. Introduction
Genomic data grants clinical researchers with
considerable prospects to study various patient
sicknesses. Nevertheless, genomic data includes a lot
of revealing information, some of which a patient
might want to remain concealed. The question from
such a patient controlled data scenario becomes: how
can an entity transact in full DNA sequence data
while concealing certain sensitive pieces of
information in the genome sequence, and yet
maintaining DNA data utility? In this paper, we
respond to this question by putting forward a
heuristic in which the patients might have a say as to
what part of their genomic information they wish to
remain concealed. The rest of the paper is organized
as follows. In Section 2, a background definition of
bioinformatics and data privacy terms is given. In
Section 3, a review of related work on the subject of
DNA privacy is done. In Section 4, the proposed
heuristic and methodology is outlined in detail. In
Section 5, the experiment and results are discussed,
and lastly in Section 6, the conclusion and future
works is given.
2. Background
DNA stands for deoxyribonucleic acid, a chemical
substance that comprises functionality information
for the make-up and growth of an organism [1]. DNA
organization is an organized collection of four
building blocks called nucleotides bases, namely
adenine (A), thymine (T), guanine (G), and cytosine
(C), in which the adenine (A) pairs with the thymine
(T), and guanine (G) with cytosine; forming a twisted
ladder-like shaped strand, referred to as a double
helix [2]. Genomics is the study of the genome
(DNA) sequence structure, the chemical components
that make up the DNA structure, including the
chemical elements that hold DNA components
together [3]. DNA sequencing refers to the process of
mapping out the nucleotide sequence of DNA [4].
Codon is a sequence of three subsequent nucleotides
in the genome that map to a specific amino acid in a
protein; a codon could also represent a triplet, which
is the start and stop points in the DNA sequence [5].
Codon frequency table or codon usage table is the
statistical representation of each codon existence in
the DNA sequence [6].
Figure 1: Google Scholar search results for “DNA Privacy”
Data obfuscation is a data privacy technique in
which data is distorted, made imprecise, and
indistinguishable so as to conceal information [7].
Data shuffling is a data privacy technique first
proposed by Dalenius and Reiss (1978) as a data
swapping methodology for categorical data [8], and

later extended by Muralidhar and Sarathy (2003) for
numerical data, shuffling is a data privacy procedure
in which sensitive data values , within an
attribute are exchanged between tuples in
the same attribute ; for instance, a sensitive value
belonging to tuple is assigned to tuple and
the value belonging to tuple is assigned to
tuple in the same attribute [9].
3. Related work
In this section, we take a look at some work being
done in the Genomics data privacy domain. Research
in Genomics data privacy is relatively still in the
early stages and not so much work and literature
exists on the subject. For instance, as illustrated in
Figure 1., a Google Scholar search results only
returned 24 documents related to the “DNA Privacy”
search term for 2013 compared to 12,700 related
documents to the search term “Data Privacy”, for
2013 [10]. However, Genomic data privacy
researchers have largely focused on using
cryptographic techniques in granting confidentiality
to DNA data sets. Malin and Sweeney (2004),
proposed employing re-identification techniques in
evaluating and modeling of anonymity protection
systems for genomic data; the goal was to prevent
any information leakage between genomic data and
named individuals in public records [11]. Malin and
Sweeney (2004)’s model focused on engineering a
secure and confidential data privacy transaction
model for genomics, without perturbing the DNA
sequence structure. However, Malin (2005), proposed
one of the first genomic data privacy models that
involved the actual manipulation of the DNA
sequence to enhance confidentiality. Malin (2005)’s
model employed k-anonymity with generalization by
ensuring that in a collection of DNA sequence data,
each generalized DNA record would appear at least k
> 1 times, to satisfy k-anonymity requirements [12].
Additionally, Malin (2005), provided a holistic
evaluation of the current state of genomic data
privacy storage and transaction systems then, by
outlining the short comings and providing
recommendations [13]. In this paper, we take a
similar approach by proposing a genomic data
privacy solution that takes into consideration both the
bio-bank architecture and the confidential transaction
of the individual’s DNA sequence. Other works have
centered on employing genomics as a tools for
cryptography [14][15]. For instance Heider and
Barnekow (2007), proposed using DNA
cryptographic and steganographic algorithms for
watermarks to detect any unauthorized use of
genetically modified organisms protected by patents
[16]. Additionally, Kantarcioglu, Jiang, Liu, and
Malin (2008) proposed a cryptographic model in
which aggregated data mining, such as, frequency
counts, could be done securely on genomic
sequences without revealing the original encrypted
raw DNA sequence [17]. On the other hand, El
Emam (2011), in a review of de-identification
technics for genomics data, observed that electronic
records are increasingly being linked to DNA bio-
banks, used by clinicians for genomic research, and
that challenges and risks still remained despite de-
identification techniques for confidentiality [15]. In
this paper, we argue for a holistic approach that
considers privacy modeling at both the bio-bank
storage level and the data transaction level between
the patient, doctor, and clinical researchers. Ayday,
Raisaro, and Hubaux (2013) proposed a holistic
model for genomics confidentiality in which data
privacy enhancing principles, using homomorphic
encryption, were employed in modeling a secure bio-
bank system, DNA data transaction process, and the
confidentiality of the DNA sequence [18].
Furthermore, Ayday, Raisaro, Hengartner,
Molyneaux, and Hubaux (2014) proposed a privacy
preserving system to grant confidentiality using
encryption to short reads of raw genomic data [19] .
In their model, Ayday et al.,(2014), suggested
storing short reads in encrypted form at a bio-bank
and then allow the stack-holders to securely retrieve
only needed short reads [19].
Figure 2: Overview of the bio-bank and DNA transaction model
From a bio-bank architectural privacy modeling
view, we follow a similar approach to Ayday et al.,
(2014) by proposing the encryption of all medical
records, including genomic data at the bio-bank level.
However, in our model, the full perturbed raw DNA
sequence, in which sensitive genomic information is
either suppressed or altered, is retrieved. In our
model, the privatized DNA sequences are retrieved
by controlling the number of queries to prevent

information leakages. For instance, a clinical
researcher in hospital A, might be interested only in a
set of genes responsible for cancer, while another
researcher at hospital B, might be interested in genes
linked to diabetes. In such scenarios, information
leaked from hospitals A and B could be used to
reconstruct sensitive information in the DNA
sequence of a patient. Another way to get around the
issue of information leakage, we propose a
publication of static tabulated results for a non-
interactive setting. In this scenario, our model would
only publish or transact in pre-defined results that
could not be changeable, after the initial obfuscation.
In this case, patients would publish the same
privatized DNA sequence to all parties involved. This
would be useful when dealing with entities such as,
insurance companies and clinical researchers, only
interested in aggregated statistical analysis.
4. Methodology
In this section we present our methodology on
obfuscation of codons with highly expressed genes,
for privacy. In this paper we assume that the DNA
data transaction will occur between the patient, data
curator (bio-bank), and a clinical researcher. The goal
is to obfuscate the DNA data so that the clinical
researchers will not view any gene information
deemed sensitive by the patient. In other words, the
patient controls by authorization, what they wish to
be disclosed. Our proposed architecture, as illustrated
in Figure 2, is composed of three components, the
main original data storage component, the
Obfuscated data component and the patient record
component. The main original data storage
component (Bio-bank): This acts as a main storage
for the patient medical records storage, including the
DNA data. The original data storage component is
highly restricted and all records are encrypted.
Access to this data storage is enforced by
documentation of who gained access, what time they
gained access, how long they accessed the system,
what information they access, and what time they
exited the storage system. In this controlled
environment, access to the storage requires another
signatory for accountability. In other words, the
system will not allow one individual to access the
system without accountability from another official.
The obfuscation data processing component: This
component, as illustrated in Figure 3, acts as a
privacy layer between the patient record access
component and the main original data storage
component. Data generation in the obfuscated data
storage is done by controlled query access on the
main original data storage component. The number
and type of queries run on the system is strictly
limited to prevent information leakage. Obfuscation
procedures are then applied to the results that are
generated by the query. The only records stored or
accessed in the obfuscated data storage component,
are those that have been sanitized (privatized) via
obfuscation and other data privacy algorithms.
Figure 3: The obfuscated data processing component
Access to the obfuscated data component and
querying process is strictly enforced using the same
security access control methods, as applied to the
original data storage component. Any data
transaction, will only involve access to the
Obfuscated data storage unit. This component is
responsible for generating obfuscated patient records
needed for any data transaction. The patient record
component: In this component, the patient and or
clinical researcher have access to the obfuscated
medical and DNA record. The patient can authorize
time restricted access to this record to their physician
or a clinical researcher. The patient record
component will document each access by a doctor,
clinician, or the patient themselves. The patient
record component could be accessible using mobile
devices, via secure and encrypted connectivity.
Phase I: In the first phase of the obfuscation, as
illustrated in Figure 3, Personal Identifiable
Information (PII) is removed from the data set. After
PII is removed, data shuffling is applied to the de-
identified data set to increase obfuscation. In our
methodology, partial data shuffling is applied on the
data set. All attributes undergo data shuffling with
exception of the Patient ID and DNA attributes. The
reason for the data shuffling is that if the controlled
and restricted data falls in the hands of an attacker, it
would be difficult to reconstruct the full identify of a
patient. Yet still, this would prevent inside attacks by

those who have minimal access to the data system at
various technical support levels. In our methodology,
a data shuffling is based on a random number
generator. The rows of data values (tuples) in each
attribute are aligned to a sequence of randomly
generated numbers, which then get sorted. The seed
for the random number generator is kept secret by the
data curators (Bio-bank), to make it difficult for any
reverse engineering. In this model, each attributed is
shuffled vertically, separately, and finally, all
attributes are combined to generate the complete
partially shuffled data set.
Figure 4: DNA data obfuscation overview
Phase II: In phase II, data obfuscation is done on
the DNA sequence, by perturbation of the codon
frequency table, as illustrated in Figure 4. In the first
step, a codon frequency table is generated for each of
the DNA sequence in the database. A frequency
analysis is then done to highlight the most frequent
codons in the table – that is, codons with the most
highly expressed genes in each amino acid group.
The perturbation by redistribution is done within the
same amino acid group. The perturbation is done on
codons with high frequency values – by redistributing
the frequencies among the codon values in the same
amino acid group. In this way, the same amino acid
could remain ‘highly expressed’ while giving a form
of obfuscation to the highly expressed genes in the
codon.
Table 1, Codon frequency for Asp before obfuscation
Asp GAT 1
Asp GAC 5
For example, the codon GAC has a high frequency
value in the aspartic (Asp) amino acid group, with a
frequency of 5 while GAT appears only once. The
total frequency for the Asp amino acid group is then
6, as shown in Tables 1.
After perturbation by redistribution of the codon
frequencies, the GAC codon gets a frequency value
of 3 while the GAT codon, a frequency value of 3.
However, the total frequency value for the Asp amino
acid group remains intact, at 6, as show in Table 2.
Table 2: Codon frequency for Asp after obfuscation
Asp GAT 3
Asp GAC 3
The totals of all codons in the frequency table stay
the same; at this stage, the DNA sequence base pair
size stays intact and the same as the original. After
perturbation by redistribution of the codon
frequencies, the new generated frequency table
(privatized table) is then used to generate a new
privatized DNA sequence. A pairwise alignment to
compare the original DNA sequence to the privatized
DNA sequence is done. A similarity values are used
to determine if an acceptable threshold value is
achieved. For example, a 50 percent similarity value
might be acceptable as threshold; if this threshold
value is not met, then a refinement is done by re-
doing the redistribution of frequency values in the
codon table. The last step is to publish the privatized
DNA sequence without revealing data from highly
expressed genes in codons with high frequencies. The
original codon frequency values are kept encrypted at
the main bio-bank storage Therefore this procedure
could be applicable by implementing the extra layer
of confidentiality on the patient DNA sequence
during data transaction from one point to another.
5. Experiment and Results
Experiment, Phase I: To test our hypothesis, we
generated a synthetic patient records data set with
150 data points containing PII information, medical
records, and the DNA sequence of each patient.
Figure 5: The random number generator seed used for shuffling
The synthetic patient records data set was used to
simulate the real original patient records data set. The
synthetic patient records data set included the
following set of attributes: {Patient ID, First Name,

Last Name, Gender, SSN, DOB, Address, City, State,
Zip code, Country, Diagnosis Code, Diagnosis
Description, PCP, and DNA}. During the first phase
of the obfuscation, the following attributes were
removed from the data set to satisfy PII sanitization
requirements: {First Name, Last Name, SSN,
Address, and City}. Data shuffling was then
individually applied to each of the following
remaining attributes: {Patient ID, Gender, DOB,
State, Zip code, Country, Diagnosis Code, Diagnosis
Description, PCP, and DNA}.
Figure 6: The values in red show the change in position after
shuffling
A random number generator was used in shuffling
the values in each separate attribute. The values were
mapped to the randomly generated numbers between
a chosen seed, in the case of our experiment, the seed
was chosen as real numbers between 0 and 1. After
the mapping the values in the attributes to the random
numbers, sorting was done. The values in the
attributes changed positions with each sorting
iteration, thus shuffling, as illustrated in Figure 6.
Figure 6, shows an illustration of a returned query
results showing the effects of shuffling on the
attribute values. The highlighted figures in red in
Figure 6, show the shift in each of the shuffled values
from their original positions.
Figure 7: Shuffling results for the date of birth attribute.
It would be difficult for an attacker to reverse
engineer the process to gain the original row of data
without prior knowledge, and considering that the
database would contain millions of data points. To
ensure that the data values in each attribute were
efficiently shuffled, we did a frequency distribution
analysis on the date of birth attribute, before and after
shuffling. As illustrated in Figures 7, the date of birth
for Patient ID P000001 was 1934 before shuffling.
However as illustrated in Figure 7, the date of birth
for the same Patient ID P000001 was substituted with
1979 after shuffling, thus offering a layer of
confidentiality for Patient ID P000001.
Experiment, Phase II: To further test our
hypothesis, we generated synthetic DNA sequences
for each of the Patient ID records in that database.
For the remainder of this paper and context of this
experiment, we shall refer to both the synthetic DNA
sequence and synthetic patient records, as original
data. Each generated original DNA sequence
contained 500 base pairs. The FASTA format was
used in formatting the raw generated original DNA
sequence. 150 original DNA sequences were
generated for the original patient database. In the
next step of the Phase II experiment, the codon
frequency of the generated original DNA sequence
was done. Codon frequency obfuscation was
achieved by redistributing values of codons with
highly expressed genes, within the same amino acid
group. For instance as illustrated in Figure 8, the
codon values for CAG and CAA in the Glutamine
(Gln) amino acid group was 2 and 9 respectively. In
the obfuscation process, the codon values for CAG
and CAA were reassigned to 5 and 6 respectively,
after redistribution.
Figure 8: Codon frequencies before and after obfuscation.
It is important to note that while the codon values
for CAG and CAA were altered, the totals for the
frequency of that particular amino acid, in this case
Glutamine (Gln), remained the same at 11. In other
words, perturbation was done to the frequency values
of the codons themselves but not the frequency
values of the amino acid group. As shown in Figure
9, the frequency of the Glutamine amino acid in the
original DNA sequence is 11, and after obfuscation,
the frequency value for the same amino acid stays the
same at 11. The obfuscated codon table was then
used to generate the obfuscated DNA sequence. A
pairwise comparison was then done on both the
original and obfuscated DNA sequences. A pairwise
analysis was done on both the original and

obfuscated DNA sequence. The goal was to
implement an obfuscation of codons with highly
expressed genes in the DNA sequence. The question
of data privacy versus utility still affects the outcome
of the results. Too much obfuscation may alter the
DNA sequence with many mutations; minimal
obfuscation could also be revealing.
Figure 9: Amino acid frequencies before and after obfuscation.
In the first part of our pairwise analysis, we
compared the original protein sequence to the
obfuscated protein sequence. Tables 3 and 4 show the
protein sequences we used for the pairwise analysis
for both the original and obfuscated data.
Table 3: Original protein sequence
>EMBOSS_001_1
QDSSGLEVAPAKEDAWYSRTMKPVQRQHP
HLGEPKRGIKVTFWIPETNPMVVLRLGSLSP
LGPISGNRRQVQSVFRRTMEQCG*DVLHL
YPTQAGPRSARRGGAQAIDN*PPCIHYGTRD
FKPSQWSSQYRVYRILP*LTNCDPPQVKPLP
LRHAVRVIM*TLRGLT
Sequence Translation tool:
http://www.ebi.ac.uk/Tools/st/
Table 4: Obfuscated protein sequence
>EMBOSS_001_1
RDAQSVPGSSTVERVLGTHDTADDVIYRT
SHLASLGRRDQYSTNEPNSSRPRSLPRTGAA
EPLLGGLQPRHYSRHPLRHGAWRKAGRNY
KRNAQVVAFNQLSLSVRSSQPFKLSRHST*Q
LFASGTGLSD*WQPHVLL*LP**LPGPH*L
GSRYRHGPCFNNQIF
Sequence Translation tool :
http://www.ebi.ac.uk/Tools/st/
As shown in Figure 10 and 11, the pairwise
alignment analysis for both the original and
obfuscated protein sequence, generated only 18
percent similarity [20]. While this might be a good
indication of privacy and obfuscation, questions
remain as to the viability of DNA data utility. The dot
matrix in Figure 11 shows regions of similarity
between the original and obfuscated protein
sequences. The x-axis represents the obfuscated
protein sequence while the y-axis shows the original
protein sequence. A diagonal line in the matrix would
represent an ideal perfect match for similarity. The
nonaligned lines represent the mismatch between the
two protein sequences.
Figure 10: The pairwise alignment between original and
obfuscated protein sequences.
Figure 11: The original and obfuscated protein sequence dot
matrix.
We then used the protein sequence of the original
DNA sequence to act as a key and reverse engineered
the obfuscated DNA sequence to obtain the original
DNA sequence by reverse translation.
Figure 12: The original and obfuscated DNA sequence dot matrix,
after reverse translation.

We used the codon frequency table of the obfuscated
DNA sequence to generate the original DNA
sequence, using the protein sequence of the original
DNA sequence as key. The dot matrix in Figure 12
shows regions of similarity between the original and
obfuscated DNA sequences after the reverse
engineering process. The x-axis represents the
obfuscated DNA sequence while the y-axis shows the
original DNA sequence. A diagonal line in the matrix
would represent an ideal perfect match or similarity.
The nonaligned lines represent the mismatch between
the two DNA sequences. As shown in Figure 12 and
13, there was a 78 percent similarity between the
original and obfuscated DNA sequences after the
reverse engineering process.
Figure 13: The dissimilarities between the original and obfuscated
DNA sequences after reverse translation.
While the focus of this paper was not to study
various reverse engineering attacks, our results show
that when the protein sequence of the original data
was used in combination with the codon frequency
table of the obfuscated DNA data, to reserve
translate, there was only a 78 percent success rate.
Figure 14: The original and obfuscated DNA sequence dot matrix.
The results could as well mean that this heuristic
was good for privacy, with 22 percent of the data
remaining concealed – relating to the obfuscation
done on codons with high frequencies and highly
expressed genes. In Figure 13, the highlighted
obfuscated DNA sequence in red, represent the
dissimilarities between the original and obfuscated
DNA sequences after the reverse translation process.
The dot matrix in Figure 14 shows regions of
similarity between the original and obfuscated DNA
sequences. The x-axis represents the obfuscated DNA
sequence while the y-axis shows the original DNA
sequence. A diagonal line in the matrix would
correspond to a perfect similarity. The nonaligned
lines represent the mismatch between the two DNA
sequences. The results in Figure 14 and 15 show the
pairwise alignment between the original DNA
sequence and the obfuscated DNA sequence
generated from the perturbed codon frequency table
without any the reverse engineering process. The
results show that there was a 50 percent similarity,
indicating that privacy might be achieved by the
concealment provided by the other 50 percent
dissimilarity; yet again, the issue of privacy versus
utility resurfaces.
Figure 15: The dissimilarities between the original and obfuscated
DNA sequences.
6. Conclusion
We have presented a codon frequency obfuscation
heuristic for genomic data privacy. DNA data privacy
in general, remains a challenge. However, data utility
remains a challenge with DNA obfuscation – more
mutations in obfuscated data. Our proposed model
shows that it might be possible to perturb the codon
frequency table by redistributing codon frequencies
of highly expressed genes within the same amino
acid group, for confidentiality. Our preliminary
results show that even with reverse engineering using
the obfuscated DNA codon frequency table in
combination with the protein sequence of the original
DNA sequence, only 78 percent was recoverable,
indicating that it might be possible to conceal certain

sections of information within the DNA sequence. In
this proposed heuristic, we envision a well informed
patient with full rights and authorization as to what
information within the DNA sequence they might
want to remain concealed. While the focus in this
paper was to present preliminary results from the
testing of the hypothesis, future works will include
testing the model against various reverse engineering
attacks. For future works include, running the same
tests using real DNA data from living organism with
large base pairs (big data). We plan on employing
other data privacy algorithms not covered in this
paper; developing a prototype of the obfuscation
architecture to automate the proposed heuristic.
7. References
[1] Genome.gov, “A Brief Guide to Genomics,”
Genome.gov. [Online]. Available:
http://www.genome.gov/18016863. [Accessed: 14-May-
2014].
[2] Dnaftb.org, “Concept 19, The DNA molecule is shaped
like a twisted ladder,” DNA From The Beginning.
[Online]. Available: http://www.dnaftb.org/19/.
[Accessed: 14-May-2014].
[3] Genomic.org.uk, “What is Genomics,” Genomic.org.uk.
[Online]. Available: http://www.genomic.org.uk/.
[4] E. B. Online, “DNA sequencing,” Encyclopædia
Britannica. [Online]. Available:
http://www.britannica.com/EBchecked/topic/422006/DN
A-sequencing. [Accessed: 14-May-2014].
[5] NIH.gov, “Codon,” Genetics Home Reference. [Online].
Available: http://ghr.nlm.nih.gov/glossary=codon.
[6] OpenWetWare.org, “Codon usage optimization,”
OpenWetWare.org/wiki. [Online]. Available:
http://openwetware.org/wiki/Codon_usage_optimization.
[7] R. Parameswaran and D. Blough, “A robust data
obfuscation approach for privacy preservation of
clustered data,” in In Workshop Proceedings of the 2005
IEEE International Conference on Data Mining, 2005,
pp. 18–25.
[8] S. E. Fienberg and J. McIntyre, “Data swapping:
Variations on a theme by Dalenius and Reiss.,” in
Privacy in statistical databases, Springer Berlin
Heidelberg, 2004, pp. 14–29.
[9] K. Muralidhar, R. Sarathy, and R. Dandekar, “Why swap
when you can shuffle? a comparison of the proximity
swap and data shuffle for numeric data.,” in Privacy in
Statistical Databases, Springer Berlin Heidelberg, 2006,
pp. 164–176.
[10] Google, “Google Scholar Search Results,” Search
Terms: “DNA Privacy”, “Data Privacy,” 2014.
[Online]. Available: http://scholar.google.com/.
[11] B. Malin and L. Sweeney, “How (not) to protect
genomic data privacy in a distributed network: using trail
re-identification to evaluate and design anonymity
protection systems.,” J. Biomed. Inform., vol. 37, no. 3,
pp. 179–92, Jun. 2004.
[12] B. A. Malin, “Protecting genomic sequence anonymity
with generalization lattices.,” Methods Inf. Med., vol. 44,
no. 5, pp. 687–692, Jan. 2005.
[13] B. A. Malin, “An evaluation of the current state of
genomic data privacy protection technology and a
roadmap for the future,” J. Am. Med. Informatics Assoc.,
vol. 12, no. 1, pp. 28–34, 2005.
[14] C. Guo, C. Chang, and Z. Wang, “A new data hiding
scheme based on dna sequence,” Int. J. Innov. Comput.
Inf. Control, vol. 8, no. 1, pp. 139–149, 2012.
[15] K. El Emam, “Methods for the de-identification of
electronic health records for genomic research.,”
Genome Med., vol. 3, no. 4, p. 25, Jan. 2011.
[16] D. Heider and A. Barnekow, “DNA-based watermarks
using the DNA-Crypt algorithm.,” BMC Bioinformatics,
vol. 8, p. 176, Jan. 2007.
[17] M. Kantarcioglu, W. Jiang, Y. Liu, and B. Malin, “A
cryptographic approach to securely share and query
genomic sequences.,” IEEE Trans. Inf. Technol.
Biomed., vol. 12, no. 5, pp. 606–17, Sep. 2008.
[18] E. Ayday, J. L. Raisaro, and J.-P. Hubaux, “Privacy-
enhancing technologies for medical tests using genomic
data,” in 20th Annual Network and Distributed System
Security Symposium (NDSS), 2013.
[19] E. Ayday, J. L. Raisaro, U. Hengartner, A. Molyneaux,
and J. Hubaux, “Privacy-preserving processing of raw
genomic data,” in In Data Privacy Management and
Autonomous Spontaneous Security, 2014, pp. 133–147.
[20] A. S.F., G. W., M. W., M. E.W., and L. D.J., “Basic
local alignment search tool,” J. Mol. Biol., vol. 215, pp.
403–410, 1990.

A Codon Frequency Obfuscation Heuristic for Raw Genomic Data Privacy

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to A Codon Frequency Obfuscation Heuristic for Raw Genomic Data Privacy (20)

More from Kato Mivule (17)

Recently uploaded (20)

A Codon Frequency Obfuscation Heuristic for Raw Genomic Data Privacy