Abstract
Genomic prediction is now a mature technology for animal breeding and genetics and many breeders’ associations and livestock, and poultry genetics companies have adopted high-through genotyping for routine breeding value prediction and for assisting in selection decisions. In this context, more research efforts in breeding and genetics are focusing more on genome annotation and phenotyping. In animal breeding, the term phenomics is almost exchangeable with high-throughput phenotyping and automated phenotyping, i.e., the recording of high dimensional phenotypic data on an organism-wide scale. High throughput phenotyping is achieved through using a variety of sensors. The sensors generate raw data that must be captured, processed, cleaned, and modeled to produce meaningful phenotypes. Animal breeders are used to working with large amounts of data. However, high throughput phenomics usually produces data that may quickly overwhelm storage and processing capabilities of even the most prepared data analysis teams. Moreover, the black-box nature of many algorithms used to process raw sensor data into actionable phenotypes usually require thorough validation and checking. Animal breeders may build on prior lessons learned processing and modeling high-throughput genotypes and traditional phenotypes and we also need to adopt new practices commonly used in the field of machine learning and computer science. Finally, high throughput phenomics will produce novel phenotypes that will require different modeling or revisiting traditional models. In this talk I will share my views on all these points, and I will illustrate them using literature data and data from my own research related to computer vision, modeling of social genetic effects and modeling of social interactions.