Criminal and Civil Identification with DNA Databases Using Bayesian Networks

Marina Andrade & Manuel Alberto M. Ferreira
International Journal of Security, (IJS), Volume (3) : Issue(4) 65
Criminal and Civil Identification with DNA Databases Using
Bayesian Networks
Marina Andrade marina.andrade@iscte.pt
Department of Quantitative Methods
ISCTE – Lisbon University Institute
Lisbon, 1649-026, Portugal
Manuel Alberto M. Ferreira manuel.ferreira@iscte.pt
Department of Quantitative Methods
ISCTE – Lisbon University Institute
Lisbon, 1649-026, Portugal
Abstract
Forensic identification problems are examples in which the study of DNA profiles
is a common approach. Here we present some problems and develop their
treatment putting the focus in the use of Object-Oriented Bayesian Networks -
OOBN. The use of DNA databases, which began in 1995 in England, has
created new challenges about its use. In Portugal the legislation for the
construction of a genetic database was defined in 2008. With this it is important
to determine how to use it in an appropriate way.
For a crime that has been committed, forensic laboratories identify genetic
characteristics in order to connect one or more individuals to it. Apart the
laboratories results it is a matter of great importance to quantify the information
obtained, i.e., to know how to evaluate and interpret the results obtained
providing support to the judicial system. Other forensic identification problems
are body identification; whether the identification of a body (or more than one)
found, together with the information of missing persons belonging to one or more
known families, for which there may be information of family members who
claimed the disappearance. In this work we intend to discuss how to use the
database; the hypotheses of interest and the database use to determine the
likelihood ratios, i.e., how to evaluate the evidence for different situations.
Keywords: Bayesian networks, DNA profiles, identification problems.
1. INTRODUCTION
The use of networks transporting probabilities began with the geneticist Sewall Wright in the
beginning of the 20th century (1921). Since then their use had different forms in several areas like
social sciences and economy – in which the used models are, in general, linear named Path
Diagrams or Structural Equations Models (SEM), and in artificial intelligence – usually non-linear
models named Bayesian networks also called Probabilistic Expert Systems (PES), [11],[14].
Bayesian networks are graphical structures for representing the probabilistic relationships among
a large number of variables and for doing probabilistic inference with those variables, [13]. Before

we approach the use of Bayesian networks to our interest problems we briefly discuss some
aspects of PES in connection with uncertainty problems in section 2.
In section 3 after presenting some possible forensic identification problems the creation and use
of DNA profile databases in some European countries is discussed putting the focus in the entry
criteria and the differences observed. How to approach evaluate and interpret the results,
whether it is a criminal or a civil identification problem is presented in section 4. Thus, it is
important to describe the Portuguese law establishing the principles to maintain a DNA database
file for civil and criminal identification purposes. The study of a criminal identification problem
considering one single perpetrator, and a civil identification problem with one volunteer, for two
different situations, are considered.
2. EXPERT SYSTEMS
Expert systems are attempts to crystallize and codify the knowledge and skills of one or more
experts into a tool that can be used by non-specialists, [14]. An expert system can be
decomposed as follows:
Expert system = knowledge base + Inference engine.
The first term on the right-hand side of the equation, knowledge base, refers the specific
knowledge domain of the problem. The inference engine is given by a set of algorithms, which
process the codification of the knowledge base jointly with any specific information known for the
application in study.
Usually it is presented in a software program, as the one we are going to show hereafter, but
such is not an imperative rule. Each of those parts is important for the inferences, but knowledge
base is crucial. The inferences obtained depend naturally on the quality of the knowledge base, of
course in association with a sophisticated inference engine. The better those parts are the best
results we can get.
A PES is a representation of a complex probability structure by means of a directed acyclic graph,
having a node for each variable, and directed links describing probabilistic causal relationships
between variables, [1]. Bayesian approach is the adequate for making inferences in probabilistic
expert systems.
2.1 Bayesian networks
Bayesian networks are graphical representations expressing qualitative relationships of
dependence and independence between variables. A Bayesian network is a directed acyclic
graph G (DAG) having a set of V vertices or nodes and directed arrows. Each node υ є V
represents a random variable Xυ with a set of possible values or states. The arrows connecting
the nodes describe conditional probability dependencies between the variables.
The set of parents, pa(υ), of a node υ comprises all the nodes in the graph with arrows ending in
υ. The probability structure is completed by specifying the conditional probability distributions for
each random variable Xυ and each possible configuration of variables associated with its parent
nodes xpa(υ). The conditional distribution of Xυ is expressed given Xpa(υ) = xpa(υ). The joint
distribution is p(x) = Π υєV p(x υ| xpa(υ)). There are algorithms to transform the network into a new
graphical representation, named junction tree of cliques, so that the conditional probability p(x υ|
xA) can be efficiently computed, for all υ є V, any set of nodes VA ⊆ , and any configuration xA of
the nodes XA. The nodes in the conditioning set A are generally nodes of observation and input of
evidence XA= xA, or they may specify hypotheses being assumed.

Software such as Hugin
1
can be used to build the Bayesian network through the graph G. That
can be done by specifying the graph nodes, their space of states and the conditional probabilities
p(x υ| xpa(υ)). In the compiling process the software will construct its internal junction tree
representation. Then, by entering the evidence XA= xA at the nodes in A, and requesting its
propagation to the remaining nodes in the network, the conditional probabilities p(x υ| xA) are
obtained. Version 6.4 of Hugin and upgrades allow the graphical use of OOBN.
OOBN are one example of the general class of Bayesian networks. An instance or object is a
regular network possessing input and output nodes as well as ordinary internal nodes. The
interface nodes have grey fringes, with the input nodes exhibiting a dotted line and the output
nodes a solid line. The instances of a given class have identical conditional probability tables for
non-input nodes. The objects are connected by directed links from output to input nodes. The
links represent identification of nodes. We use bold face to refer the object classes and math
mode to refer the nodes. The modular flexibility structure of the OOBN is of great advantage in
complex cases
3. Forensic identification problems
The use of DNA profiles in forensic identification problems has become, in the last years, an
almost regular procedure in many and different situations. Among those are: 1) disputed paternity
problems, in which it is necessary to determine if the putative father of a child is or is not the true
father; 2) criminal cases as if a certain individual Y was the origin of a stain found in the scene of
a crime; or in more complex cases to determine if an individual or more did contribute to a mixture
trace found; 3) civil identification problems, i.e., the case of a body identification, together with the
information of a missing person belonging to a known family, or the identification of more than
one body resultant of a disaster or an attempt. And even immigration cases in which it is
important to establish family relations.
Here the focus is to approach the civil and criminal identification problems. The establishment
and use of DNA database files for a great number of European countries worked as a motivation
to study in more detail the mentioned problems and the use of these database files identification.
The use of a DNA profile database may allow delinquents’ identification and/or the connection of
criminal conducts and the respective individuals, the exclusion of innocents as well as the
recognition and civil identification of missing people. A genetic profile database may be an
important help in forensic investigation, particularly in crimes of repetitive tendency, when DNA
samples of condemned individuals are collected. In the context of the civil identification it may be
very useful when unidentified corpses appear and may be identified by comparison of their DNA
profiles family volunteer’s profiles.
3.1 DNA database files
The discovery of biological fingerprints in 1984 opened new perspectives in forensic identification
area. Since then the technical advances and results obtained have allowed studying and solving
increasingly complex problems. Almost twenty five years after Alec Jeffreys’ team discovery,
Portugal established the legislation for the construction of a genetic database, law nº5/2008.
The advances in DNA technology and knowledge opened new perspectives for civil and criminal
investigation. Apart from the ethical and legal questions that are in the domain of the legal
system, we should draw some attention to the experience and knowledge acquired by those
countries that already have their own databases operating and try to learn from them ways to
improve on how to operate with the Portuguese database, [5].
1
http://www.hugin.com - OOBN a resource available in the Hugin 6.4 software.

TABLE 1: National DNA databases in Europe.
There has been a considerable discussion about the individuals to include in a DNA proﬁles
database, specially with different results in countries with different legal systems. The main
differences are linked to the emphasis given by the countries: to the individuals or the social
order.
In the table above (TABLE 1) it is possible to observe significant differences between the
European countries in what concerns criteria to enter a person into a database (while a suspect
or only after conviction, different types of conviction). The criteria to remove records and the
number of entries in the database also have important differences all over Europe.
As we have seen there are clear differences between countries more on the north and more on
the south of Europe, which is essentially due the different perspective valuing more the social
order or the individual itself.
4. Criminal and civil identification using DNA profile databases
Let us consider a criminal case in which a DNA profile has been recovered from a crime scene,
and it is admitted to be left by the culprit (only one perpetrator); and a civil identification
problem with one volunteer giving his/her genetic information.
The Portuguese law n°5/2008 establishes the principles for creating and maintaining a database
of DNA profiles for identification purposes, and regulates the collection, processing and
Country Year Entry criteria for suspects (convicted offenders)
England 1995 Any recordable offence
Austria 1997 Any recordable offence
Croatia/Switzerland 2000 Any recordable offence
Germany 1998 > 1 year in prison (after court decision)
Finland 1999 > 1 year in prison
Denmark 2000 > 1.5 years in prison
Norway 1999 Many serious offences (after court decision)
Hungary 2003 5 years in prison
Sweden 2000 No suspects entered (> 2 years convicted)
Belgium 2002 No suspects entered (after court decision)
Netherlands 1997 No suspects entered (> 4 years convicted)
France 2001 No suspects entered (serious offences, voluntary samples only)
Spain 1998 Phoenix program – civil database for civil ident. vol. donations
Portugal 2008 Vol.,“problem samples”, “reference samples” (≥ 3 years convicted)
Italy - Law in preparation

conservation of samples of human cells, their analysis and collection of DNA profiles, the
methodology for comparison of DNA profiles taken from the samples, and the processing and
storage of information in a computer file.
Here it is assumed that the database is composed of a file containing information of samples from
convicted offenders with 3 years of imprisonment or more - α ; a file containing the information of
samples of volunteers - β ; a file containing information on the "problem samples" or "reference
samples" from corpses, or parts of corpses, or thing or place where the authorities collect
samples - γ .
4.1 Criminal identification – one single perpetrator
For a crime that has been committed, forensic laboratories identify genetic characteristics in order
to connect one or more individuals to the crime. Apart from the laboratories results it is a matter of
great importance to quantify the information obtained, i.e., to know how to evaluate and interpret
the results obtained providing a support to the judicial decision. Here it is assumed a DNA trace
found at a scene of a crime left by only one perpetrator.
“The experience with some databases seams to indicate that before the commitment of a serious
crime, some suspects have already been involved in minor offences. This fact associated to the
repetitive motif of some serious crimes can support the importance of DNA databases not only to
the criminal investigation but also to the prevention of crime, mainly if there are large entry
criteria.”, [4].
Some notation:
Let cC be the genetic characteristic (DNA profile) found at the crime scene, and sC the
suspect’s genetic characteristic. The evidence is ( )sc CCE ,= , the DNA typing results for the
crime sample and the suspect. Our interest is to discuss how to evaluate the hypotheses and the
odds ratio. In court the hypotheses are: HP: The suspect (s) left the crime stain vs HD: Some other
person left the crime stain.
There is a match between the crime scene profile and the suspect’s profile. The court wants to
compare the two preceding hypotheses. It is important to discuss the presentation of the
evidence in court and how to evaluate the hypotheses of interest. The posterior odds are:
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )α
α
α
α
α
α
α
α
α
α
∈
∈
∈
∈
=
∈
∈
∈
∈
=
∈
∈
sHP
sHP
sHCCP
sHCCP
sHP
sHP
sHEP
sHEP
sEHP
sEHP
D
P
LR
Dsc
Psc
D
P
D
P
D
P
|
|
,|,
,|,
|
|
,|
,|
,|
,|
444 3444 21
The likelihood ratio, LR, takes the form:
( )
( )
( )
( )α
α
α
α
∈
∈
∈
∈
=
sHCP
sHCP
sHCCP
sHCCP
LR
Ds
Ps
Dsc
Psc
,|
,|
,,|
,,|

Whether or not the suspect left the crime sample that does not provide any information to our
uncertainty about his/her genetic characteristic or genotype, i.e.,
( ) ( )αα ∈=∈ sHCPsHCP DsPs ,, . Therefore
( )
( )
.
,,|
,,|
α
α
∈
∈
=
sHCCP
sHCCP
LR
Dsc
Psc
In a case with a trace of a single perpetrator, for each marker, one can use the object-oriented
Bayesian (OOBN) network shown in Figure 1. Each node (instance) in the network represents
itself a Bayesian network. Nodes spg and smg are all of class founder, a network with only one
node which states are the alleles in the problem and the respective frequencies in the population,
and represent the suspect’s paternal and maternal inheritance. Node sgt and cgt are of class
genotype. The genotype of an individual is an unordered pair of alleles inherited from paternal
(pg) and maternal (mg) genes, here represented by gtmin:=min{pg, mg} and gtmax:=max{pg,
mg}, where pg and mg are input nodes identical to the gene node of founder. The nodes cmg and
cpg specify whether the correspondent allele is or is not from the suspect. If s=c? has true for
value then the true perpetrator’s allele will be identical with the suspect’s allele, otherwise the true
perpetrator’s allele is chosen randomly from another man in the population. The single node s=c?
represent the binary query ‘Is the suspect the perpetrator?’
FIGURE 1: Network for a criminal case with a single perpetrator.
Here, if the suspect, s, is in the file of convicted offenders then ( ) ( )αα ∈>∈ sHPsHP DP
|| ,
otherwise those probabilities may be assumed equal. Therefore, the posterior odds are:
If s is in the file of convicted offenders Otherwise
( )
( )
( ) ( )
.
,,|
1
|
|
,,|
1
1
αα
α
α ∈
>
∈
∈
∈
>
sHCCPsHP
sHP
sHCCP DscD
P
Dsc
44 344 21

We need to assess ( )α∈sHP P
| and ( )α∈sHP D
| . The posterior odds computation is in the
judges’ domain, we can only explain how to do it and how to interpret the evidence, i.e., how to
use it as a decision support element.
4.2 Civil identification – one missing person and one volunteer
A missing individual is reported. A body is found. The hypotheses are: HP: The body found is the
body of the claimed individual X vs HD: The body found is any other individual’s, not the claimed
individual X. People want their relatives alive and reject the hypothesis that states their lost
relative is dead. A volunteer supplies genetic material to be used in the test of a partial match.
The evidence is ( )volBf CCE ,= - the genetic characteristics of the body found, BFC , and the
volunteer, volC . The posterior odds is
( )
( )
( )
( )
( )
( )γβ
γβ
γβ
γβ
γβ
γβ
,|
,|
,,|
,,|
,,|
,,|
∈
∈
∈
∈
=
∈
∈
volHP
volHP
volHEP
volHEP
volEHP
volEHP
D
P
D
P
D
P
One may assume ( ) ( )γβγβ ,|,| ∈=∈ volHPvolHP DP then
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
.
,,|
,,|
,,,|,
,,,|
,,|
,,|
,,|,
,,|,
,,|
,,|
1
4444 34444 21
44444 344444 21
=
∈
∈
∈
∈
=
∈
∈
∈
∈
=
∈
∈
γβ
γβ
γβ
γβ
γβ
γβ
γβ
γβ
γβ
γβ
volHCP
volHCP
volHCCP
volHCCP
volEHP
volEHP
volHCCP
volHCCP
volEHP
volEHP
Dvol
Pvol
DvolBF
PvolBF
D
P
LR
DvolBF
PvolBF
D
P
The conditioning does not include the information of the body found. Whether or not the volunteer
is related with the individual whose body was found that does not provide information to our
uncertainty about his genotype.
Depending on the volunteer and the claimed individual family relation we only may observe a
partial match with one volunteer. Apart form that, it is important to check if there is a match
between BFC and any of the “problem samples” inγ . Assuming no match of BFC and any sample
inγ , the likelihood ratio may be written as:
( )
( )
.
|
,|
DBF
PvolBF
HCP
HCCP
In a case having only one volunteer the likelihood ratio can be computed using a Bayesian
network. Suppose we have a missing individual, an elder person and a son or a daughter
claiming the disappearance and who gives a genetic profile voluntarily. The likelihood ratio can be
computed using the following network:

FIGURE 2: Network for civil identification with one volunteer, paternal or maternal relationship case.
As above nodes ancpg and ancmg are all of class founder, a network with only one node which
states are the alleles in the problem and the respective frequencies in the population, and
represent the volunteer’s ancient paternal and maternal inheritance. Node volgt and bfgt are of
class genotype, the volunteer and body found genotypes.
The nodes tancmg and tancpg specify whether the correspondent allele is or is not from the
volunter. If bf_match_vol? has true for value then the volunteer’s allele will be identical with the
volunteer’s allele. The node bfancg defines the Mendel inheritance in which the allele of the
individual whose body was found is chosen at random from the ancient’s paternal and maternal
gene.
For a case with only one volunteer, but now for example a brother or a sister of an individual who
is missing and is being searched, the likelihood ratio may be computed using the following
Bayesian network:
FIGURE 3: for civil identification with one volunteer, brother or sister relationship case.

Nodes ancpmg and ancppg are all of class founder and represent the volunteer’s ancient genes
of paternal and maternal inheritance. That inheritance will pass to nodes volpg and volmg, which
are going to form volunteer genotype. Node volgt and bfgt are of class genotype, the volunteer
and body found genotypes. The remaining nodes are the same as the ones presented in the
previous problem.
5. CONCLUSION & FUTURE WORK
To connect an individual with a crime on the basis of a profile match may be dangerous because
the database may contain undetected errors. In order to avoid misclassification with DNA from
the database it is important to admit, at least, a second and independent analysis.
After computing the likelihood, whether it is a criminal case or a civil identification case, it is
possible to compute the posterior odds, i.e., multiplying the likelihood ratio and the prior odds, in
order to perform a comparative evaluation between the prosecution and the defense hypotheses.
The database file α is a subset of the population set P, P⊂α . If the size of the database file is
small, then one may only have a small fraction of the possible offenders. Therefore, it is important
to take that into account. This topic should be considered in future work.
Whether it is criminal or civil identification in many situations the evidence may have more than
one individual involved. In future work that must be considered. Also, in civil identification
problems an important issue is to study how to compute the likelihood ratios when there is a
match or a partial match between the genetic characteristic of the individual whose body was
found and the file of “problem samples” and “reference samples”, γ .
6. REFERENCES
1. A. P. Dawid, J. Mortera, V. L. Pascali, D. W. Van Boxel. “Probabilistic expert systems for
forensic inference from genetic markers”. Scandinavian Journal of Statistics, 29:577-595,
2002
2. B. P. Battula, K. Rani , S. Prasad , T. Sudha. ”Techniques in Computer Forensics: A
Recovery Perspective”. International Journal of Security, Volume 3, Issue 2:27-35, 2009
3. David J. Balding. ”The DNA database controversy”. Biometrics, 58(1):241-244, 2002
4. F. Corte-Real. “Forensic DNA databases”. Forensic Science International, 146s:s143-s144,
2004
5. G. Skinner. “Multi-Dimensional Privacy Protection for Digital Collaborations”. International
Journal of Security, Volume 1, Issue 1:22-31, 2007
6. I. Evett and B. S. Weir. “Interpreting DNA Evidence: Statistical Genetics for Forensic
Scientists”, Sinauer Associates, Inc. (1998)
7. M. Andrade, M. A. M. Ferreira. “Bayesian networks in forensic identification problems”.
Journal of Applied Mathematics. Volume 2, number 3, 13-30, 2009

8. M. Andrade, M. A. M. Ferreira, J. A. Filipe. “Evidence evaluation in DNA mixture traces”.
Journal of Mathematics and Allied Fields (Scientific Journals International-Published online).
Volume 2, issue 2, 2008
9. M. Andrade, M. A. M. Ferreira, J. A. Filipe., M. Coelho. “Paternity dispute: is it important to be
conservative?”. Aplimat – Journal of Applied Mathematics. Volume 1, number 2, 2008
10. M. Guillén, M. V. Lareu, C. Pestoni, A. Salas and A. Carrecedo. “Ethical-legal problems of
DNA databases in criminal investigation”. Journal of Medical Ethics, 26:266-271, 2000
11. M. N. Anyanwu, S. Shiva. “Comparative Analysis of Serial Decision Tree
Classification Algorithms”. International Journal of Computer Science and Security,
Volume 3, Issue 3:230-240, 2009
12. P. Martin. “National DNA databases – practice and practability. A forum for discussion”. In
International Congress Series 1261, 1-8
13. R. E. Neapolitan. “Learning Bayesian networks” , Pearson Prentice Hall, (2004)
14. R. G. Cowell, A. P. Dawid, S. L. Lauritzen, D. J. Spiegelhalter. “Probabilistic Expert Systems”,
Springer, New York, (1999).

Criminal and Civil Identification with DNA Databases Using Bayesian Networks

More Related Content

Similar to Criminal and Civil Identification with DNA Databases Using Bayesian Networks (20)

Recently uploaded (20)

Criminal and Civil Identification with DNA Databases Using Bayesian Networks