Managing abdominal fat in broilers is a constant challenge in modern poultry production. Excess fat not only reduces carcass quality and meat yield, but also represents a significant waste of feed, directly impacting the efficiency and profitability of poultry operations. Traditionally, selecting birds with lower abdominal fat has been a slow and costly process, as it often requires the slaughter of animals in order to obtain precise measurements.
However, in September 2025, the scientific paper “Optimization of Genomic Breeding Value Estimation Model for Abdominal Fat Traits Based on Machine Learning” was presented, proposing an innovative solution that combines genomics with machine learning to predict with unprecedented accuracy the genetic predisposition of a bird to accumulate abdominal fat, opening new and efficient avenues for genetic improvement.
Overcoming the limitations of the past
Traditional breeding methods rely on pedigree records and phenotypic measurements (observable traits), which is impractical for traits such as abdominal fat. The advent of genomic selection (GS) represented a major breakthrough, making it possible to predict an individual’s genetic value from markers in its DNA. Even so, the enormous volume of genetic data available has made it necessary to develop more sophisticated methods to identify which markers are truly important.
A two-stage genetic selection strategy
The research team developed a framework that optimises the selection of genetic markers (known as SNPs) to subsequently feed them into an advanced prediction model.
- Initial identification of key genes: First, a genome-wide association study (GWAS) combined with linkage disequilibrium (LD) analysis was used. This technique functions as a primary large-scale filter, enabling the identification of genomic regions most strongly associated with abdominal fat accumulation while discarding thousands of irrelevant markers.
- Refinement with machine learning: A two-step strategy was then applied to further refine the selection:
- Reduction with Lasso: A machine learning model called Lasso dramatically reduced the number of candidate genetic markers, retaining only those with the most significant influence.
- Optimisation with RFE: Subsequently, another algorithm called Recursive Feature Elimination (RFE) reviewed this selection to obtain a final set of genetic markers with maximum predictive power.
This dual-filtering process made it possible to identify a set of 177 key genetic markers that proved to be highly predictive of abdominal fat in the reference population.
DAWSELF: an intelligent and dynamic prediction model
With the most informative genetic markers already selected, the researchers developed a novel machine learning system called DAWSELF (Dynamic Adaptive Weighted Stacking Ensemble Learning Framework).
Rather than relying on a single model, DAWSELF combines several different prediction models, both linear and non-linear. Its main innovation is that it dynamically assigns a “weight” or level of importance to each model based on its performance. It is like having a committee of experts where the opinions of the most accurate carry greater weight. This multi-layer “stacking” approach allows the system to learn from data in a far more robust and precise manner than any individual model could achieve on its own.
The DAWSELF framework can be adapted to predict other economically important traits such as:
Disease resistance (coccidiosis, salmonellosis)
Egg quality (weight, shell colour, albumen height)
Animal welfare (skeletal health, viability)
Environmental adaptation (heat stress resistance)
Integration with emerging technologies
Combining DAWSELF with other technologies can further enhance its utility:
IoT sensors for automated phenotyping
Artificial intelligence for image analysis
Blockchain for genetic traceability
Gene editing for functional validation of candidate genes
Although DAWSELF demonstrated excellent performance in the populations studied, poultry geneticists must consider population-specific validation for their particular genetic lines. Differences in population structure, selection history and environment may affect the transferability of predictive models.
Results: exceptional accuracy validated in the field
The results are compelling. The DAWSELF framework, using the refined set of 177 genetic markers, achieved a predictive accuracy of 99.65% in the reference population.
More importantly for its practical application, the system was successfully validated in three independent populations, including the AA commercial line. In these populations, DAWSELF maintained an extraordinarily high accuracy, consistently exceeding 97.9% and demonstrating its robustness and generalisability.
Practical applications in poultry production
This technological advance has direct and highly valuable implications for poultry production:
- Early and precise selection: It enables poultry genetics companies and breeding enterprises to select the best breeding stock at a very early stage, without the need for post-mortem measurements. This greatly accelerates genetic progress towards leaner lines.
- Improved feed efficiency: By selecting birds genetically predisposed to lower abdominal fat deposition, the amount of feed “wasted” on producing unwanted fat is reduced. This translates into an improvement in the feed conversion ratio, one of the cornerstones of poultry profitability.
- Increased economic returns: Lower fat deposition and a higher proportion of lean meat in the carcass improve the quality of the final product and the economic return for the producer.
- Advanced genetic advisory services: Geneticists and production veterinarians must maintain continuous feedback and work alongside artificial intelligence (AI) experts for the implementation of genomic selection programmes based on these new machine learning tools, helping the poultry industry to become even more competitive and efficient.
The application of Artificial Intelligence to poultry genetics will lead to progressively superior commercial poultry strains.
The integration of genomics and artificial intelligence, as demonstrated by the DAWSELF framework, represents a qualitative leap in poultry breeding. It provides a tool of exceptional precision to address a complex and economically important trait such as abdominal fat. For the poultry sector, this means the possibility of producing higher-quality broilers in a more sustainable and profitable manner, consolidating the role of technology as a fundamental pillar in the animal production of the future.
Source:
-. Optimization of Genomic Breeding Value Estimation Model for Abdominal Fat Traits Based on Machine Learning. Chen, H.; Dou, D.; Lu, M.; Liu, X.; Chang, C.; Zhang, F.; Yang, S.; Cao, Z.; Luan, P.; Li, Y.; et al. . Animals 2025, 15, 2843. https://doi.org/10.3390/ani15192843