Talavera I., Silva F., Hernández N., González R., Palau J., Ferreira M. M. C., "Application of Chemometrics Tools for Automatic Classification and Profile Extraction of DNA Samples in Forensic Tasks". Águas de Lindóia, SP, Brazil, 10-15/09/2006: 10th International Conference on Chemometrics in Analytical Chemistry (CAC-2006, CAC-X), Book of Abstracts (2006) OP09. Oral 09.

10th International Conference on Chemometrics in Analytical Chemistry OP09

Application of Chemometrics Tools for Automatic Classification
and Profile Extraction of DNA Samples in Forensic Tasks

Isneri Talaver¹*, Francisco Silva¹, Noslén Hernández¹, Ricardo González¹, Juan Palau¹,
Márcia M. C. Ferreira²   italavera@cenatav.co.cu

¹Advanced Technologies Application Center. MINBAS. Havana, Cuba. ²Instituto de Química, Universidade
Estadual de Campinas, Campinas – SP, 13083-970 BRAZIL

Keywords: Classificaiton, Support Vector Machine, Image Analysis
_____________________________________________________________________________________
 

    DNA profiling has attracted a  good deal  of  public attention in recent years.  The  practical application of
DNA  technology  to  the identification  of  biological material  has  a  significant  impact  on forensic biology,
because it enables much stronger conclusions in genetic identity tasks¹.  During laboratory data generation,
the forensic scientist conducts experiments to transform the biological samples into observable  DNA  data².
To carry out this transformation our forensic especialists applies Electrophoresis Analysis on Polyacrilamide
Gels with silver tintion reagent, and the DNA sequences are visualized as black spots on the gels.  There is
a standardized method to manually detect spots  of  DNA  and make the number designations,  but it is very
tedious, and inefficient way to do the task.  In this paper an automatic solution is presented which integrates
image processing, and pattern recognition techniques.
    After digital image acquisition and image  preprocessing³  the next step is  the description of the spots.  A
representation using  14  boundary and regions descriptors was chosen.  To find out which  of them are the
most significant to characterize DNA spots,  a  combination of  a  PCA  analysis  and  a  C4.5 Decision Tree
were used.  All the spots present on the polyacrilamide gel images  were described automatically,  using the
most significant features obtained.  For the profile extraction only DNA spots were useful;  therefore,  it  was
necessary  to  solve a two class classification problem among  DNA  spots  and  No-DNA  spots.  In order to
perform  the   classification   process   with   high   velocity,   effectiveness   and   robustness,   comparative
classification studies amongSupport Vector Machine, K-NN  and  PLS-DA  classifiers were made.  The best
results obtained with  the  SVM  classifier  demonstrated the advantages attributed to it in the literature as a
two class classifier⁴.
    After the Classification process,  an image with  only  DNA  spots was obtained. For the profile extraction,
first  it  was  necessary  to  determine   the  regions  and  sub regions  in  the  image  that  contain  the DNA
sequences patterns  (STR Loci). These patterns contain all the possible alleles present in a population with
especific numbers  and  it  is  possible to visualize  them in the image as sequences of  DNA black spots for
each STR Loci.  It is necessary to use a set of  12 STR Loci in ordeer to be able to conform the profiles. To
solve the task,  the first  step  was  the detection  of  the  candidate's  regions  according  to  the  intensities
histogram along the x  axis;  second the determination inside of these regions of  the  periodic  sequence of
the image according the characteristics  of  the  patterns and then we finished  with the determination of the
sub regions  applying   a   Sequential  Cluster  Leader  Algorithm⁵.  Sometimes,  as   a   consequence  of  a
malfunction  of  the classifiction algorithm,  or by difficulties in the electrophoresis chemical process,  one or
more spots inside a sequence of  SRT Loci  pattern  were  missing;  to  restore  them a  new  algorithm was
developed. The process finalizes with the profile extraction of the experimental DNA samples.  To solve this
task,  the formula  for  distance  of one point  to  a  straight  line was applied.  The numbers assigned to the
experimental spots were the same assigned  to  the lines of the patterns whose distances  are the minors to
them.
    A set of original plates was processed  by  the expert using  the  standardized manual procedure and the
results of the profile extraction were compared with  the results obtained applying the automatic method.  A
success rate of  97%  and  a  significant  decrease in the time's response,  indicated that this method has a
very good  computational behavior,  effectiveness,  and  provides  a  very  useful tool  to  reduce  time  and
increase  the  quality  of  the forensic specialist responses.  A  software which implements all of the method
was developed.
__________________________________________________________________________________________________________________________

References

¹ Gill U.; Millican E.; Oldroyd N.; Watson S.; Sparkers. Advances in Forensic Haemogenetics, 1996; 6:235.
² Weber J.; May P. Am. J. Hum. Genet. 1989:44; 388-96.
³ Silva F.; Talavera I.; Gonzáles R.; Hernández N.; Palau J.; Santiesteban M. LNCS 3773, 2005; 242-251.
⁴ Nianyi.; Wecong L.; Jie Y.; Guozheng L. Support Vector Machine in Chemistry. World Scientific
Publishing Co. 2004.
⁵ Hartigan J. Clustering Algorithm. John Wiley and Sons. New York 1975.