AG2PI Field Day #4 - January 20, 2021

Event was rescheduled for January 27, 2021

In traditional evaluations (ones not using DNA data) the model contains an effect for management group which is cows in the same herd calving at a similar time.

The 30 cattle chromosomes are divided into over 600 segments. Imputation gives us the alleles for each of the animal's 2 haplotypes for each segment. If we never find an animal that has the same haplotype from both parents and the haplotype frequencies indicate we should, we assume that there is a lethal that prevents the birth of a live calf. We confirm our hypothesis by seeing if particular ratings of carrier bulls to daughters of carrier bulls have depressed fertility. We assume that the homozygous state causes early embryonic death because we have not received reports of abortions from these matings.

See Lifetime net merit for details

Yes, an increase in inbreeding is a consequence of selection. The effects of inbreeding depression are countered by detection of deleterious recessives and use of mating programs to restrain the level of inbreeding on a DNA basis.

The job of an evaluation system is to decompose the observation into the genetic component of interest and the other factors. Sometimes data is adjusted prior to the evaluation, other times the effect is included in the model.

There is not a near term plan to do genomic evaluation for SCR. ScCR has been described as a phenotypes measure because so many other factors beside genetics determine the success of a unit of semen. There is interest in determining if there are any variants in the full genome associated with sire fertility. If significant markers are found we could report on them.

Thanks for the reference. Multi-trait analysis is a powerful technique, we use it routinely in the evaluation of type traits, and is some other cases it is under consideration.

As fluid milk has become a decreasing portion of the use of milk and more has gone to cheese and other manufactured produce, the volume of milk becomes a factor in the cost of hauling the milk rather than a valuable component.

Currently data warehouses move towards data lakes which are more fit for ml approaches

THAT IS ONLY FOR MILK RECORDINGS, RESEARCH DATA IS OFTEN GONE. The other interesting phenotypes such as Dry matter intakes from earlier studies, Metabolic profiles, etc are not maintained.

Thank you for your question! We have used deep learning algorithms for semantic segmentation. The two main algorithms we have used are 1) Mask R-CNN and 2) U-Net.

Thank you for your question! We started our experiments using RFID to identify animals. We still use it sometimes. However, our decision to start using imaging technology to identify animals was based on several challenges imposed by RFID systems. First, in the long-term, computer vision systems will become a more reliable technology for animal traceability. RFID technology is still prone to error and fraud compared to animal biometry. Second, the precision level related to animal location, when housed in group, is not enough to differentiate two animals standing together (short-range) on the feed alley. One cannot tell if both animals are eating or if one animal is eating and the other is only standing with the head out of the headlock. The third aspect is related to optimized usage of data streams. Even if you identify the animal, you still need images or other sensors to predict behavior; thus you need to collect data from an additional source. Using image analyses, you can predict behavior and identification through the same data source from a single device. For sure computational cost goes up as you decide to go with image instead of RFID, but opportunities for new and feasible tools also increase. Besides, our vision is that the development of new technologies for phenotyping should be focused on optimized devices (multi-task) as opposed to single-task sensors.