BIF 2019: The use of 'Big Data' in a modern swine breeding program now and in the future

Jeremy Howard
Smithfield Premium Genetics 

Smithfield Premium Genetics is the nucleus that provides sows (over 1 million) for Smithfield and the sires of the terminal market pigs. At Smithfield, they mate a Landrace to a Large White to produce the commercial F1 sow. These sows are then breed to a Duroc. The terminal pig then has maximal heterosis (maternal heterosis and direct heterosis). SPG uses single sire semen of Durocs to mate to commerical farms in Missouri and North Carolina. They collect 60,000 carcass data points per year.
On the maternal side, it is had to get stayability data because generations are turned over so quickly. They use commercial test herds to collect this data on sows.
Howard said that genomic information on purebred animals prior to selection allows them to better predict performance in a commercial setting. This genomic data also allows them to figure out if problematic meat is produced at a company owned farm or an outside source.

Big data is driven by volume and speed at which the data comes in. Their big data is based on pictures, sensors, and sound data, which has minimal human intervention during data collection.

The use of electronic feeder systems use RFID tags to determine which pig is at the feeder, amount of feed consumed, and the body weight of the animal. In a single year there are 2 million animal observations per year in the SPG electronic feeder systems. They use robust regression to identify outlier data points (feeder is empty and has no feed, scale is miscalibrated, etc.). Key piece is for farmers to have a dashboard that alerts them to issues.

They have also collect line speed packing plant data collection.

Research is working on using images and machine learning to classify whether or not the carcass from a pig has had it tail bitten.

Researchers also use a camera to estimate the volume, length, heights and roundness to estimate the body weight of pigs. Right now, these predictions are accurate to within 10 pounds.

Researchers are also working on facial recognition in pigs. If animals have markings on their face, facial recognition is easy. However, for pigs with white faces, facial recognition is much harder. Also, can you track the animal as it ages.

For new technology to be used, researchers have to prove it is useful for genetic prediction and improves the profitability of an operation.

Howard believes big data will make it easier to obtain relevant phenotypes at the commercial level. It's important to not that in swine breeding, all the nucleus animals are genotyped. What is needed is collecting data on their commercial offspring.


Comments

Popular posts from this blog

ARSBC 2018: EPDs and Reasonable Expectations in Commercial Crossbred Operations

Neogen launches upgraded Igenity® Beef Profile

EPDs 101: Use Information to Improve Your Herd