Validation of genomic predictions and genomic reliability Mel
Validation of genomic predictions and genomic reliability Mel Tooker and Paul Van. Raden USDA, Agricultural Research Service, Animal Genomics and Improvement Laboratory, Beltsville, MD, USA melvin. tooker@ars. usda. gov CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 1) Mel Tooker
Topics Interbull validation of genomic predictions (GPTAs) – Predict later deregressed GPTA from earlier GPTA, weighted by later genomic reliability (GREL) Simpler validation of GPTA – Predict later GPTA from earlier GPTA Simple validation of GREL – Estimate earlier GREL from later GREL, genetic standard deviation (SD), and SD of change (later GPTA – earlier GPTA) Gains in reliability (REL) from more frequent updates CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 2) Mel Tooker
Interbull International Bull Evaluation Service (Uppsala, Sweden) Permanent subcommittee of the International Committee for Animal Recording (ICAR) Responsible for: – Documenting evaluation systems – Forming technical workgroups – Multiple-trait, across-country evaluation (MACE) Selling domestic semen in many other countries requires that methods used to calculate PTA and GPTA are validated by Interbull every 2 years CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 3) Mel Tooker
2017 GPTA Trend validation – ideal net merit (NM$) 900 800 700 600 500 400 300 200 100 0 0 100 200 300 400 500 600 700 800 2014 GPTA CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 4) Mel Tooker
2017 GPTA Trend validation – NM$ for proven bulls 900 800 700 600 500 400 300 200 100 0 0 100 200 300 400 500 600 700 800 2014 GPTA CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 5) Mel Tooker
2017 GPTA Trend validation – NM$ for all bulls Ideal Proven bulls Young bulls 900 800 700 600 500 400 300 200 100 0 0 100 200 300 400 500 600 700 800 2014 GPTA CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 6) Mel Tooker
Holstein GPTA validation (preliminary) Trait Milk Fat Protein Somatic cell score Daughter pregnancy rate Heifer conception rate Final score Strength Slope* 0. 98 0. 92 0. 88 1. 06 1. 04 1. 03 Intercept* – 113 – 2. 4 – 1. 7 0. 3 0. 1 – 0. 3 – 0. 2 0. 1 *After base change adjustment CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 7) Mel Tooker
Simple GPTA validation results 3, 984 young Holstein bulls from Aug. 2014 with >100 daughters in Aug. 2017 Trait Net merit Milk Fat Protein Productive life Somatic cell score Daughter pregnancy rate Cow conception rate Heifer conception rate R 2 (%) 82 81 81 80 83 77 79 79 68 Intercept* – 12 – 131 – 3 – 0. 04 0. 1 0. 4 Slope* 1. 00 1. 01 0. 95 0. 94 1. 07 1. 01 1. 06 1. 08 0. 97 *After base change adjustment CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 8) Mel Tooker
Top proven Holstein bulls (August 2014) Now with >1, 000 daughters NM$ Bull Robust Erdman Twist Alta. Greatest Alta. Fairway Diesel Yano Facebook Awesome Manifold Top 10 average 2017 744 704 551 644 571 457 466 440 492 558 563 *Base-adjusted values 2014* 615 593 567 555 550 522 518 504 501 543 CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 9) PA 2014* 339 304 249 304 376 302 328 320 322 260 311 Progeny (2017) Daughters AI sons 2, 319 102 5, 433 10 2, 302 4 1, 365 10 2, 460 0 1, 242 0 11, 074 14 5, 766 14 7, 334 0 58, 030 3 9, 733 16 Mel Tooker
Top young Holstein bulls (August 2014) Now with >1, 000 daughters NM$ Bull Troy Rogers Cabriolet Ponder Emerald Bombero Halogen Jayden Supersire Donatello Top 10 average 2017 635 688 866 726 512 693 323 626 855 755 668 *Base-adjusted values 2014* 747 708 680 660 656 648 642 642 639 666 CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 10) PA 2014* 494 533 507 533 500 538 475 464 510 561 511 Progeny (2017) Daughters AI sons 1, 209 58 1, 338 17 6, 354 10 1, 295 6 1, 198 6 1, 245 35 1, 922 83 2, 927 3 16, 627 260 4, 743 18 3, 886 50 Mel Tooker
Top proven Jersey bulls (August 2014) Now with >100 daughters NM$ Bull Volcano Magnum Link Daybreak Hickey Bindy Arhil Zimpel Victory Memo Top 10 average 2017 433 501 427 510 429 395 346 408 351 158 396 *Base-adjusted values 2014* 515 497 450 410 405 404 402 396 385 383 425 CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 11) PA 2014* 313 218 213 281 292 153 149 190 295 – 64 204 Progeny (2017) Daughters AI sons 4, 167 29 5, 985 10 1, 760 3 935 5 266 4 619 0 106 0 100 0 1, 522 2 243 0 5 1, 570 Mel Tooker
Top young Jersey bulls (August 2014) Now with >100 daughters NM$ Bull Harris Mackenzie Machete Formidable Hector Walter Revolution Marlo Pilgrim Chili Top 10 average 2017 654 525 421 424 393 561 457 684 559 514 519 *Base-adjusted values 2014* 567 536 535 530 528 522 516 508 486 522 CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 12) PA 2014* 386 313 379 405 376 389 347 420 298 335 365 Progeny (2017) Daughters AI sons 2, 113 59 383 8 2, 283 8 642 6 138 1 289 3 215 2 407 19 600 4 984 17 805 13 Mel Tooker
Data for validating GREL Published GPTAs – April 2014 (GREL 2014 ) – April 2017 (GREL 2017 ) SD of difference in GPTAs REML estimates of SD of true transmitting ability (TA) from Interbull MACE CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 13) Mel Tooker
Example GREL validation (Holstein protein) Average published GREL 2014 was 0. 76 GREL 2017 was 0. 95 SD of change was 8. 4 REML TA SD was 17. 5 Observed GREL 2014 for protein calculated as: GREL 2014 = 0. 95 – 8. 4 2/17. 5 2 = 0. 72 CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 14) Mel Tooker
Observed vs. published GREL (2014) Trait* Milk Fat Protein PL SCS DPR NM$ Average Jersey Obs Pub 73 68 72 68 71 68 47 55 64 62 63 52 68 64 65 62 Diff +5 +4 +3 -8 +2 +11 +4 +3 Obs 72 74 72 65 77 69 68 71 Holstein Pub 76 76 76 70 73 68 73 73 Diff – 4 – 2 – 4 – 5 +4 +1 – 5 – 2 *PL = productive life; SCS = somatic cell score; DPR = daughter pregnancy rate CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 15) Mel Tooker
Average REL for NM$ by age 100 90 80 Bulls in AI Females REL (%) 70 60 50 40 30 20 10 0 <1 1 CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 16) 2 3 Age (years) 4 5 6 Mel Tooker
Phenotypic update frequency Suppose REL increases steadily from REL 1 to REL 2 across a year Gain in REL from n updates per year (REL n) instead of 1 annual update should average: Example: Suppose average bull REL increases from 75% (REL 1) to 91% (REL 2) when 4 years old (no daughters → many daughters) – Minimum gain is 0% with an annual update because bulls would stay at 75% for the whole year – Maximum gain is 8% with instant updating; bulls would average (75 + 91)/2 = 83% during that year CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 17) Mel Tooker
Phenotypic update frequency CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 18) (continued) Mel Tooker
REL gains by update frequency Frequency Annual 6 months 4 months 3 months 2 months Monthly Weekly Daily Instant Updates 1 2 3 4 6 12 52 365 ∞ Young Marginal REL (%) gain 73. 10 0. 00 73. 70 0. 60 73. 90 0. 20 74. 00 0. 10 74. 10 0. 10 74. 28 0. 08 74. 29 0. 01 74. 30 0. 01 CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 19) Proven Marginal REL (%) gain 75. 00 0. 00 79. 00 4. 00 80. 30 1. 30 81. 00 0. 70 81. 60 0. 60 82. 30 0. 70 82. 80 0. 50 82. 97 0. 17 83. 00 0. 03 Mel Tooker
Conclusions Simpler validation of GPTA is easier to compute and explain, but not quite as independent New procedure developed to validate GREL GPTA properties are very close to expected GRELs were slightly too high (2%) for Holsteins, slightly too low (3%) for Jerseys Recent GPTAs (young and old) may all be too low (genetic trend is underestimated) Genetic progress is fast! CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 20) Mel Tooker
Acknowledgements Interbull Working Group on Genomic Reliability (Zengting Liu, Paul Van. Raden, Martin Lidauer, Mario Calus, Vincent Ducrocq, Haifa Benhajali, and Hossein Jorjani) Council on Dairy Cattle Breeding for DHI data from dairy farmers USDA-ARS project 1265 -31000 -101 -00, “Improving Genetic Predictions in Dairy Animals Using Phenotypic and Genomic Information” CDCB Industry Meeting, Madison, WI – October 3, 2017 ( 21) Mel Tooker
- Slides: 21