engleski

Artificial intelligence in future animal selection

Recent development of high throughput genomic molecular technologies has provided platform for implementation of genomic selection for affordable price in numerous livestock breeding programs. However, most economically important traits exhibit a very complex genetic architecture and by being very difficult to identify individual causal loci, prediction of BVs for these traits, even in the era of genomics, prevalently rely on statistical techniques that typically assume the infinitesimal genetic model. In addition to the advancements of high throughput genomic technologies, a tremendous progress has been made with technologies designed for “real time” monitoring and recording of animals and farms. Specially designed cameras, sensors, scales, counters and other purposely developed devices (e.g. electronic estrus detection devices, artificial noses, near infrared spectrometers etc.) provide platform to effectively collect valuable information for affordable price and in less laborious way than before. However, there is general impression among the members of the animal science community that the methodology to simultaneously analyze this overwhelming amount information somehow lags behind. It seems that huge amount of information remains unexploited to the full potential. I would dare to say that methodology exists, but it is used by very few people, and only in tackling some very specific issues. It is the artificial intelligence (AI) and its’ subfield machine learning (ML) which is dedicated to the development of algorithms for prediction and inference. There are essentially two major types of ML, supervised (when inputs and desired output are available) and unsupervised (when only inputs are available). In the phase of “learning” or constructing models, ML aims to choose from a pool of candidate probability models that can best predict unobserved data. The full potential of ML algorithms in the “precision animal agriculture” has not been investigated and exploited enough so far. AI has been used to some extent in some advanced livestock husbandry facilities by analyzing the data collected by sensors and other hardware technologies and by providing solutions by mimicking human decision- making (e.g. open the gate if some criteria is met). However, there is still a plenty of room for its usage, especially in the field of animal breeding and selection. This scientific field, due its inherent characteristics, is expected to benefit the most from the implementation of AI in the near future. Many scientists hypothesize that ML algorithms designed to “learn and conclude” from diverse, complex and high- dimensional data could at least to some extent tackle numerous unresolved issue in this field. Their hypothesis are usually based upon the potential possibility of ML algorithms to capture nonlinear dependencies and unknown interactions across multiple variables which somehow correspond with the non-additive (epistatic and dominant) gene effects. ML algorithms, especially that known as deep learning (multi-layer) algorithms were proved to be a powerful tool to make accurate predictions from complex data in other scientific fields, but there is very few information about their power and ability to predict phenotypic values from molecular data. Many questions still remain to be answered here, but numerous ongoing studies kind of promise that AI and ML will play an important role on this issue in the very near future.

Machine learning, animal breeding, selection

nije evidentirano