Developing innovative digital technology and genomic approaches to
livestock genetic improvement in developing countries
Raphael Mrode, Julie Ojango, John Gibson and Okeyo Mwai
12th World Conference on Animal Production (WCAP)
Vancouver, Canada, 5-8 July 2018

Structure of most genetic Improvement Program in developed
countries
Data
Collection
and storage
(PI)
Analytical tools (PPP)
Breeding Programs or Delivery of
appropriate genetics to farmers
(PI)
Herd management information
Genetic evaluations Farmers
Foreign Genetics
Good Foundation for sustainability or continuous genetic progress

Structure of most genetic Improvement Program in developing
countries
Breeding Programs or
Delivery of appropriate
genetics to farmers (PI)
Farmers
Foreign Genetics
Indigenous
breeds or
crosses
No Foundation for sustainability or continuous genetic progress
No Training
No feedback

0
1000
2000
3000
4000
5000
Harsh Poor Good
Yield(l)
Production environment
Indigenous
X-bred
Exotic
Indigenous (non-dairy)
breeds: highly adapted to
harsh conditions; little
potential to increase milk
yield under better
feeding.
Crossbreds: Already exist
in lager numbers &
respond to better feeding
with increased milk yield;
moderately well adapted.
Exotic dairy breeds: very
high genetic potential for
milk yield which is NOT
expressed under the most
favourable conditions;
poorly adapted
Typical low-input
smallholder environment
Success comes from appropriately matching the
genetic potential with production environments,
(physical, management and markets)

What Challenges are being addressed by ADGG?
little or no systematic and
sustainable breeding
programs
limited access to the dairy
genetics or breed type that best
suit the different production
systems
Farmers do not have access to
various services, esp. AI & health,
and inputs necessary to support
sustained productivity
improvement
Inadequate access to
input & market
services
access to information or farmer
education and training services
lacking so cant improve herd
productivity and system profitability

What is ADGG
Platform for African Dairy Genetic Gains (ADGG) is a multi-country,
multi-institution ILRI-led pioneering proof of concept which is:
developing and testing a multi-country genetic
gains platform that uses on-farm performance
information and basic genomic data for identifying
and proving superior crossbred bulls for AI delivery
and planned natural mating for the benefit of
smallholder farmers in Africa

ADGG: Approach and Objectives
Innovative application of existing &
emerging technologies
1. To establish National Dairy Performance Recording Centers
(DPRCs) for herd and cow data collection, synthesis, genetic
evaluation and timely farmer-feedbacks
2. To develop & pilot an ICT platform (FFIP) to capture herd, cow
level & other related data & link it to DPRCs (feeds back key
related herd/cow summaries, dairy extension & market info.
etc.).
3. To develop low density genomic chip for breed composition
determination & related bull certification systems for
crossbred bulls

Pr
o
d
uc
ti
vi
ty
Optimize Realized
Productivity for
farmers
+
Smart application of
Genomic tools
Engagement (PPP &
Capacity
development)
Effective phenotyping
Dairy cattle: Innovative application
of Genomics & ICT thro PPPs

Cow
Calendar
AI/Animal Health
Technician
Genomic
chip
Increased
efficiency and
profitability
Platform
Platform

Detailed data
DB
(cleaned)
Hand-
held
gadget
Error
Log
Data
querying
Analytics
Samples
R P
y
F
Field staff
Consultation
Analyzed
data/samples
DB
(results per
script)
Visualized results
etc.)
Further Data cleaning
PPP
Feedback Farmers
Other
scientists
External AccessOther primary data
(e.g., service
providers)
Clear MTAs
are key!Clear MTAs
are key!

What data is being collected?
• Milk yield (litres on test-day basis) & milking frequency
• Lactation length (derived)
• Calving interval (derived)
• Birth weight
• Calf/heifer growth (weight for age)
• Survival to different age (derived)
• AFC (derived)
• Lactation number
• Type and incidence of disease, interventions/treatment & costs
• Body condition score on the test –day
• Teat and udder type scores, including disorders
• Behavioral (movement/rumination) tracking in association
estrus detection
• Milk composition (provided for but not being measured now)
• Fat %
• Protein %
• Urea (mg/dl)
• Lactose %
• Somatic cell count (cells/ml)
The above are being used to develop breeding objectives

Some results

Tanzania sites Ethiopia sites

• Metabase
(http://metabase.data.ilri.org/
dashboards)
• Summaries prepared on basis of
request from each partner

Raw statistics on test-Day milk records from Tanzania
Row Labels
Total number
of TD Records
Avg Test-Day
milk ± SD
Max
TD-Milk
Min TD-
Milk
Number of animals
with TD Records
Arusha 5103 7.27±3.49 25 0.5 1624
Iringa 2322 5.99±2.71 20 0.5 817
Kilimanjaro 5325 6.57±4.06 28 0.5 2050
Mbeya 6377 9.51±4.41 30 1 2260
Morogoro 1 9.00 9 9 1
Njombe 3225 10.06±4.00 28 0.5 1016
Tanga 5837 6.23±3.76 25 0.5 2178
Grand Total 28190 7.64±4.18 30 0.5 9946

Data collection - Sustainability
• Several options are being explored
• Training individuals to undertake a bundle of services:
• AI services, supply inputs to farmers and data collection or
verification
• Easy to use phone based ICT tools
• farmers can use to input data. Data is only verified by personnel
providing bundle services occasionally during farm visits.
• well supported with very good feedback that farmers can see
benefits of data collection

Characteristics of genomic data in small
holder systems
• Genotyping activities undertaken as a result of research
project or breed associations
• Challenges:
• Number of genotyped animals are few and are mostly females
• Reference are mostly females but a few males in some occasions
• Very difficult to clearly define data sets for validation and
therefore these are created either by random or structured
sampling
• Mostly on cross breeds animals

Characteristics of genomic data in small
holder systems
• Opportunities:
• Genotypes on cross breed animals
• Explore novel methods adapted to the data structure
• Plurality of Bayesian methods have been examined

Quick wins from genomics in small holder
systems
• With lack of systematic data and pedigree, quick
wins from genotypic data includes
• Reduces the need for accurate pedigree recording as
genomic relationship can easily be computed
• Parentage discovery using SNP data
• Determining breed composition
• Future use in traceability of animal products

Experimenting with DGEA data
• Examine genomic prediction using the DGEA small holder data
• Phenotypes were milk yield deviations on 1034 cows (Kenya)
with HD genotypes
• Exotic breeds were Friesian, Holstein, Guernsey, Jersey and
Ayrshires and indigenous breeds : Nellore, Zebu
• Four classes of animals created: cows with > 87.5%, 61−87.5%,
36−60%, and < 36% exotic breed genes.

Classification of cows by breed
composition
Dairy% Breedtype Dairy%by-Breedtype
1 (0-20%) Zebu 6 1
2 (0.33-35%) Mixed + Zebu 6 2
3 (36-60%) Ayr/Gue/Jer 1 3
3 Fri 2 4
3 Mixed + Zebu 6 5
4 (61-87.5%) Ayr 1 6
4 Fri 2 7
4 Ayr/Fri 3 8
4 Gue/Fri 4 9
4 Ayr/Gue/Fri 5 10
4 Mixed + Zebu 6 11
5(>87.5%) Ayr 1 12
5 Fri 2 13
5 Ayr/Fri 3 14
5 Gue/Fri 4 15
5 Gue/Fri/Ayr 5 16

Developing a low density SNP Chip for
breed composition
• Low density snp assay can be developed that gives accurate
estimates of dairy proportion

Low density assay for breed composition
• Based on DGEA data
• A low density assay of about 200 SNPs have been
identified for determination of breed composition
• If parentage verification is included, assay expands to
400 SNPs
• Piloting its utilization among farmers for certification
of bulls, heifer and cows.

Experimenting with DGEA data
• Non-Bayesian method: GBLUP & SNP-BLUP
• Bayesian methods
• Determine possible levels of accuracy for young bulls of
different breed compositions with no data

Current status : Accuracies of genomic
prediction –based on DGEA data
Validation
Population
Validation set Reference Set Accuracy
>87.5% Exotic 297 716 0.41
60 – 87.5%
Exotics
448 565 0.35
50 -33 % Exotics 178 835 0.32
A. Brown,* J. Ojango,† J. Gibson,‡ M. Coffey,* M. Okeyo,† and R.
Mrode*†1
*†
Short communication: Genomic selection in a crossbred cattle
population using data from the Dairy Genetics East Africa Project

Looking at other models
 GBLUP with G matrix and a dominance matrix (D) :
 Multi-trait approach.
 Multi-trait GBLUP - proportion of exotic and indigenous genes
as separate effects.
 GEBV was computed for each of 9 chromosome regions for
top cows with high exotic or indigenous genes.
 Regions with highest contribution to GEBV of top cow were
 chromosomes 14, 8, 1 and 2 for cows with high exotic genes
 chromosome 3,1,10 and 5 for cows with high indigenous
genes

Accuracy of genomic prediction
0
0.1
0.2
0.3
0.4
>87.4
61-87.5
36-60
>36
GBLUP GBLUP + Dominance
GBLUP based
on breed
proportion
GBLUP based on
breed proportion
(correlated)

Regression Coefficients
GBLUP GBLUP + Dominance
GBLUP based
on breed
proportion
GBLUP based on
breed proportion
(correlated)
0
0.5
1
1.5
2
2.5
3
3.5
4
>87.4
61-87.5
36-60
>36

 Results from the multi-trait approach are encouraging but we need to
understand it more
 Idea is to find the haplotypes or chromosome segments whose origin
can be traced to exotic and indigenous breeds . These can then be fitted
separately
 This is being explored by admixture within chromosome in collaboration
with UNE

Incorporating genotypes from foreign sires
 Li et at 2016 examined the improvement in prediction
reliabilities for 3 production traits from multi-trait SSGBLUP
 Brazilian Holsteins that had no genotypes
 adding information from Nordic and French
Holstein bulls that had genotypes.
 Interbull proofs for foreign countries used as
correlated traits
 Some average increases in reliabilities with foreign
information included for bulls were
 Milk : 2% ; Fat : 45% and none for protein

Other methods explored
• Currently from GBLUP or SNP-BLUP top cows tended to be
dominated with cows of high degree of exotic genes.
• Ideally, you want improve the milk production in cows with
about medium indigenous genes
• This therefore prompted using a weighted G in GBLUP.
• Gwt was computed using SNP effects from BayesA and BayesB
• SNPs effects were from all data, exotic (> 0.65%), indigenous
(<= 0.65%) or combined exotic and indigenous (0.25:0.75)

Other methods explored
• Accuracy of predictions only changed for cows with <36%
exotic genes
• Accuracies were 0.12, 0.16 , 0.21 and 0.21 for all data,
exotic, indigenous and combined respectively
• Ranking of indigenous cows in the top 40% increased by 14%
with using SNP effects from indigenous or combined

Results cont..
Variable measure At start (2016) Now (2018)
Number genotyped
to inform genomic
selection
0 10000
Genomically
evaluated x-bred
breeding bulls
0 By July 2018

Delivery of improved genetics
 Carvalheiro (2014) described some business models for
Nellore Cattle in Brazil
 One of the models: GEBVs are predicted by multinational
private company and are regarded as intellectual property of
the company
 breeders or the breeding programs do not have access to the
genotypes and genomic prediction equations but rely on
company selling GEBVs

Delivery of improved genetics
 Developed countries
 AI - companies and breed societies play major roles in
 Developing countries
 Farmer cooperatives
 National Artificial Insemination Centers (NAIC)
 AI usage is low
 Bull rankings from Data Centers must be linked to NAIC or
farmer cooperative
 Genotyping strategy must ensure key bulls using for natural
mating in villages are part of the evaluation systems
 Foreign AI companies should support data collection to
include daughters of their bull

• Unstable internet connectivity but
we are able get around it!
• Weak institutions
• Inadequate/not fully supportive
policy to PPR-arrangements
Some Challenges

• The proof of concept has been quite
successful and the 1st batch of genomically
evaluated crossbred bulls expected by end
of this year
• Given data structure for smaller , well
adapted methodologies will be needed for
genomic selection and initial results are
encouraging
• In order to benefit from large numbers and
diverse genetic backgrounds, regional
programs is the way to go
Conclusions

• Need to collaborate with developed countries
where some of the sires of these cows could be
genotyped
• Parentage discovery
• Genotype by environmental interaction if enough data
• Huge opportunities exist for overlaying
other studies to the ADGG’s framework
with potentially exciting results and
outcomes
Conclusions

Dairy Farmers & Farmer
organizations
National/regional
Institutions/govts.
Acknowledgements
The organizers of the Big
Data Conference

This presentation is licensed for use under the Creative Commons Attribution 4.0 International Licence.
better lives through livestock
ilri.org
ILRI thanks all donors and organizations who globally supported its work through their contributions
to the CGIAR system