Predicting Energy BalanceStatus of Holstein Cows usingMid-Infrared Spectral Data

Sinéad Mc Parland,G.Banos, E.Wall, M.P.Coffey, H.Soyeurt,

R.F.Veerkamp & D.P.Berry

Introduction Energy balance (output-input) is a heritable

indicator of health & fertility in dairy cows Useful for multi-trait breeding programme BUT

Expensive to measure (correctly) Measurement not feasible on commercial herds Little data available

Methods to model energy balance exist Require expensive phenotypes Rely on phenotypes not always available

Example of Energy Balance Prediction

Milk fat content Milk protein content

:Predicted Energy Balance

1 2

3

Potential errors

0.0

0.2

0.4

0.60.8

1.01.2

1.4

1.61.8

2.0

200 400 600 800 1000 1200 1400

Pin Number

MIR

read

ing

Objective

Predicted Energy Balance

•Predict energy balancedirectly from milk usingMIR spectral data

•Can we improve theaccuracy of prediction? 0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

200 400 600 800 1000 1200 1400

Pin Number

MIR

read

ing

Materials and Methods1. Data Collection

Langhill experimental herd of Holstein cows (SAC,Scotland) Two genetically divergent lines Two feeding systems

Routinely recorded phenotypic traitsMilk, fat, protein, DMI, live weight & BCS

Random regressions fit to get daily solutions Fixed effects: experiment group, year-season of calving,

calving age, year-by-month of record Random effect: cow*Σ(DIM)Models fit within parity Data retained between 1990-2010

Materials and Methods2. Calculation of energy balance Two separate measures (Banos & Coffey, 2010)

Direct_EB = inputs – outputsincl. milk production, DMI, weight, BCS & diet

Body energy content (EC) = predicted protein andlipid weights from BCS and LWT

ALSO

Daily deviation from mean direct_EB (dev_EB)Cows own deviation within parity

Materials and Methods3. Mid Infrared Spectral (MIR) data

Monthly samples from all cows sent for MIRanalysis September 2008 – December 2009 Light shone through each milk sample 1,060 wavelength readings for each sample

Materials and Methods3. Mid Infrared Spectral (MIR) data

Monthly samples from all cows sent for MIRanalysis September 2008 – December 2009 Light shone through each milk sample 1,060 wavelength readings for each sample

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

925 1425 1925 2425 2925 3425 3925 4425 4925

Wavelength

MIR

read

ing

Materials and Methods4. Prediction equations

Partial least squares analysis (PROC PLS, SAS)

Two models – MIR onlyMIR + milk yield

AM, PM & MD yields analysed separately1,199 AM, 1,127 PM and 1,148 MD records available

Cross validation method (max 20 factors) Also external validation

25% of data set independently tested Best model has the highest R2 for EXT. validation

RESULTS

Energy Balance Lactation Curves

-70

-60

-50

-40

-30

-20

-10

0

10

20

5 25 45 65 85 105

125

145

165

185

205

225

245

265

285

305

Days in Milk

Ener

gyBa

lanc

e(M

J/da

y)

Parity 1 Parity 2 Parity 3

Energy Balance - Feed Group

-70

-60

-50

-40

-30

-20

-10

0

10

20

5 25 45 65 8510

512

514

516

518

520

5225 24

5265 285 30

5

Days in Milk

Ener

gyBa

lanc

e(M

J/d

ay)

High Concentrate Low Concentrate

Energy Content Lactation Curves

4000

4500

5000

5500

6000

6500

7000

5 55 105 155 205 255 305

Days in Milk

Ener

gyCon

tent

(MJ) Parity 1 Parity 2 Parity 3

Cross Validation Results

10210.38DEV_ EB1611290.24Energy Content12270.32Direct_EB

PM16210.37DEV_ EB1611440.23Energy Content16260.35Direct_EB

MD17200.40DEV_ EB1711310.25Energy Content18250.41Direct_EB

AM

FactorsRMSER2

Addition of milk yield as a predictor

0.440.38DEV_ EB0.240.24Energy Content0.420.32Direct_EB

PM0.410.37DEV_ EB0.220.23Energy Content0.430.35Direct_EB

MD0.440.40DEV_ EB0.250.25Energy Content0.500.41Direct_EB

AMMIR & YieldMIR onlyPredictors

Update

Data collection on-going

Since collation of results presented, data

size (MIR) has doubled

Analyses re-run

Results updated -

0.390.480.38DEV_ EB0.200.380.24Energy Content0.450.530.32Direct_EB

PM0.400.470.37DEV_ EB0.190.360.23Energy Content0.440.470.35Direct_EB

MD0.390.450.40DEV_ EB0.180.340.25Energy Content0.420.430.41Direct_EBR2R2R2AM

ExternalCrossCrossValidationNew ResultsPrevious Results

Conclusion

Predicting energy balance directly from milk ismore accurate than using fat:protein ratio

Greater predictive ability when milk yield includedin the model

New data aided improved predictive ability Predictive ability for external validation <50%

Still a lot of unexplained variation “Noisy” phenotype as measured here

Work on-going to improve equations

This work was carried out as part of theRobustMilk project that is financially supportedby the European Commission under the Seventh

Research Framework Programme, GrantAgreement KBBE-211708

www.robustmilk.eu

Acknowledgements