Content uploaded by Natacha Nikolic
Author content
All content in this area was uploaded by Natacha Nikolic on Apr 11, 2014
Content may be subject to copyright.
What do Artificial Neural Networks tell us about the genetic structure of
populations? The example of European Pig populations.
Natacha NIKOLIC a, Young-Seuk PARK b, Magali SANCRISTOBAL a, Sovan LEK c and Claude CHEVALET a
MS
PI
DU
LANDRACE
National population
Commercial line
MW
BI
BL
LB
MA
TA
LS
PR
PU
BB
GO
BK
LARGE WHITE
National population
Commercial line
CA CT
CS
NC
RE NI
BS AS
MJ
HA
HA
NS
CR
TM
LE
DR
LA
IR
IR
0
0,1
0,2
0,3
0,4
0,5
0,6
0,4 0,45 0,5 0,55 0,6 0,65 0,7
National populations
Commercial lines
LANDRACE
Iberian breeds
British Lop
Icelandic Landrace
DUROC
HAMPSHIRE
PIETRAIN
LARGE WHITE MEISHAN
LANDRACE
British Lop
Icelandic Landrace
LARGE WHITE
MEISHAN
PIETRAIN
DUROC
HAMPSHIRE
Iberian breeds
SOM cluster 5
a Laboratoire de Génétique Cellulaire (UMR 444), INRA, BP 52627, 31326 Castanet Tolosan Cedex, France
b Department of Biology, Kyung Hee University, Dongdaemun-gu, Seoul 130-701, Korea
c Laboratoire Evolution de la Diversité Biologique (UMR 5274), CNRS, 118 route de Narbonne, 31400 Toulouse Cedex 4, France
Nikolic et al. (2009) Genet. Res. Camb. 91, 121-132
Acknowledgements
This research was based on the results gathered in the PigBioDiv European
project (BIO4 CT 98 0188), which is gratefully acknowledged. 1
ALIMENTATION
AGRICULTURE
ENVIRONNEMENT
Objective
Assess the added value of the unsupervised approach given by Artificial Neural Networks (Self-Organizing Map), compared with classical
approaches.
0.1
DEBB01
GBPI04
DEPI03
FRPI02
FRPI05
SELS01
DELR14
FRLR01
ITLR03
FRLR13
DKLR05
NOLR08
DKLR04
FILR06
GBLR12
GBBL01
GBLR10
GBLR11
CNMS01
ITCT01
DEDU03
GBDU02
ITDU01
ESNC01
ISLR09
DELW02
ITLW03
DELW10
FRLW01FRLW12
FRLW08 FRLW09
GBLW07
GBLW05
GBLW06
PLPU01
CZPR01
DEAS01
FRCR01
GBTA01
GBBS01
GBBK01 GBGO01
PTBI01
DEHA02
GBHA01
ITCA01
ITCS01
ITNS01
DEMA01 ESMJ01
ESNI01
ESRE01
GBLB01
GBMW01
Large White
(LW)
Duroc
(DU)
Landrace
(LR)
Piétrain
(PI)
Hampshire
(HA)
Chinese
Meishan
(MS)
Data: PigBioDiv data (http://www.projects.roslin.ac.uk/pigbiodiv/.): 50 unlinked microsatellites, 60 populations, giving 2737 individuals and 700 alleles
The Neighbor-Joining tree in SanCristobal et al. (2006a)
clustered lines within main breeds in a star like manner. Large
branch length mostly represent a large effect of genetic drift.
Method
SOM (Kohonen 2001) is an unsupervised learning algorithm, that performs a nonlinear projection of multivariate data onto lower dimension.
Use of the SOM toolbox in Matlab developped by the Laboratory of Information and Computer Science in the Helsinki University of Technology (Alhoniemi et al 2000)
A hexagonal grid of 10x20 ells gave a good topographic error rate
equal to 5%.
The map was classified into 8 clusters
The limits between clusters are visualized with the dark points.
1
1.5
2
2.5
3
3.5
4
4.5
5
67
5
2
48
3
1
LW PI MS
LR HA
DU
The individuals of these breeds
are clustered together (assigned
to a single SOM cell or to
neighboring cells).
The individuals of some of the
synthetic lines and of some
breeds are spread in different
locations
Expected heterozygosity
Relative SOM diversity
LW LR PI DU HA MS
The Principal Component Analysis separated the Chinese Meishan (MS) from Western breeds, and
displayed clusters of lines within main breeds. Some overlap between clusters is to be noted.
Bayesian clustering of all individuals with Structure: choosing the number of clusters may be difficult with large complex data, and the convergence of the chain sometimes problematic
Conclusions
- Kohonen’SOM provided a global view of the data without any prior hypothesis on
their organisation.
- Many individuals can appear on a descriptive finite space
- The nonlinear projection gave useful information on the complex organisation of the
global diversity, by clustering similar individuals and then spreading them from such a
cluster without overlap between clusters
The relative SOM (within breed) diversity for main breeds
is related to the expected heterozygosity
