Anette van Dorland ILRI, Addis Ababa, Ethiopia, 26 February 2003 Clustering of breed types: Preliminary results.

Anette van Dorland

ILRI, Addis Ababa, Ethiopia, 26 February 2003

Clustering of breed types: Preliminary results

Introduction

1. Large number of unknown breed types:

How different/similar are these breed types from each other ?

2. Farmers knowledge versus enumerator observation

Multivariate techniques

Introduction (cont.)

Approach I:

Grouping of entities based on the multivariate

similarities among the entities

No prior information of the formed groups available

Cluster analysis

Approach II:

Grouping of entities based on the multivariate similarities

among the entities

Prior information of the formed groups available

Discriminant analysis

Data on cattle from Borana Zone

Five woreda’s selected (see map)

Borana Zone

Bore

HagereMariam

Liben

Dire

Teltele

Three woreda’s predominantly in

lowland (Dire, Liben and Teltele)

Two woreda’s predominantly in

highland (Bore and Hagere Mariam)

Borana Zone

Oromia Region

209 records on breed types

26 qualitative variables on phenotypic characteristics

First step: Principal Components Analysis

Second step: Agglomerative Hierarchical Clustering (AHC)

Mahalanobis’ distance (dissimilarity)

Strong linkage as aggregation criteria

Data and Methodology

Principal Components Analysis

Characteristic

Coat colour-body

Coat pattern

Hair sizeHair type

Frame sizeDewlap sizeHump sizeHump shapeFace profile

Back profileRump profileEar sizeEar shapeEar orientationHorn lengthHorn shapeHorn orientationHorn spacingTail lengthUdder sizeTeat sizeNavel flap size

Coat colour-headCoat colour-ears

Coat colour-hoofCoat colour-tail

10 principal componentsresponsible for 64 % of the variation between the observations

Principal Components Analysis (cont.)

Characteristic PC1 PC2 PC3Body colour 1 22 1 0Body colour 2 0 0 7Head colour 20 1 0Ear colour 21 1 0Tail switch colour 5 2 10Hoof colour 9 3 4Coat colour pattern 1 1 0Hair length 2 3 13Hair type 0 1 2Frame size 2 8 0Dewlap size 1 15 0Hump size 0 10 6Hump orientation 0 0 6Face profile 1 1 17Back profile 1 2 0Rump profile 1 0 8Horn shape 0 0 1Horn orientation 0 0 2Horn spacing 1 1 7Horn length 0 3 1Ear size 0 8 0Ear shape 2 0 0Ear orientation 3 2 12Tail length 0 4 0Udder size 3 11 1Teat size 2 10 1Navel flap size 2 13 1TOTAL 100 100 100

Contributions of the variables (%)

Agglomerative Hierarchical Clustering: Dendrogram

0.0

1.0

2.0

3.0

4.0

5.0

Dissimilarity

Dendrogram

Dendrogram (cont.)

4.3

4.4

Dissimilarity

Cluster 1

Cluster 2

Cluster 3

(11 observations)

(70 observations)

(128 observations)

Distribution of animals of cluster 1

Distribution of animals of cluster 2

Distribution of animals of cluster 3

Coat colour of body: cluster 1

0

5

10

15

20

25

30

35

40

1 2 3 4 5 6

Coat colour combination of body

% o

f h

ou

seh

old

s

Coat colour of body: cluster 2

0

5

10

15

20

25

1 2 3 4 5 6

Coat colour combination of body

% o

f h

ou

seh

old

s

Coat colour of body: cluster 3

0

5

10

15

20

25

1 2 3 4 5 6

Coat colour combination of body

% o

f h

ou

seh

old

s

Physical characteristics

Phenotypic characteristic Cluster 1 Cluster 2 Cluster 3Frame size Short 0 31 28

Medium 45 53 53Long 55 16 19

Dewlap size Absent 0 0 2Small 9 26 46Medium 73 61 46Large 18 13 6

Hump size Absent 0 0 1Small 0 49 63Medium 91 49 35Large 9 3 1

Physical characteristics (cont.)

Phenotypic characteristic Cluster 1 Cluster 2 Cluster 3Face profile Flat 0 97 83

Convex 9 1 9Concave 91 1 8

Ear orientation Erect 40 9 11Lateral 60 56 82Drooping 0 36 6

Udder size Small 0 23 46Medium 90 57 45Large 10 20 9

Navel flap Absent 0 7 27Small 30 34 53Medium 60 50 15Large 10 9 5

Distribution of clusters by agro-ecological zone

0

10

20

30

40

50

60

70

80

Dega Weina Dega Kolla

% o

f b

ree

d t

yp

es

Cluster 1 Cluster 2 Cluster 3

Distribution of clusters by production system

01020304050607080

Crop-livestocksystem

Agro-pastoralists Pastoralists

% o

f b

ree

d t

yp

es

Cluster 1 Cluster 2 Cluster 3

Quality of traits: Production traits

0102030405060708090

100

% o

f h

ou

se

ho

lds

Cluster 1 Cluster 2 Cluster 3

Quality of traits: Adaptation traits

0

10

20

30

40

50

60

70

80

90

Diseasetolerance

Droughttolerance

Ability to walklong distances

% o

f h

ou

seh

old

s

Cluster 1 Cluster 2 Cluster 3

Suggestion

4.3

4.4

Dissimilarity

Cluster 1

Cluster 2

Cluster 3

‘Borana’ group

‘Guji’ group

?

Distribution of breed types (farmers’ knowledge)

Borana Zone

Breed type

Guji

Arsi

Borana

Konso

Ogaden

ArsixBorana

BoranaxGuji

BoranaxKonso

Unknown

Further analysis…..

0.0

1.0

2.0

3.0

4.0

5.0

Dissimilarity

Dendrogram

Conclusions

Multivariate techniques can be used for on-farm breed

characterization work by classifying the observations on

individual animals into well-defined breed types/strains

Multivariate techniques can help formulating hypotheses,

which can be tested using detailed genetic studies

Multivariate techniques can

facilitate more focused genetic

studies including molecular

biology