Chromosomal screening of candidate genes associated with F2 chicken birth weight using the random forest algorithm

Genomic selection has substantially improved animal breeding by enabling the early identification of genetically superior animals. This advancement is particularly important in poultry breeding, where production efficiency is highly dependent on growth traits such as birth weight. However, the high cost of high-density SNP arrays has stimulated interest in developing cost-effective low-density panels without compromising genomic prediction accuracy. In this study, we employed the Random Forest (RF) algorithm to identify informative SNPs for the development of low-density SNP panels for genomic prediction of birth weight in broilers. Genotypic data from an F2 population derived from reciprocal crosses between Arian broilers and Urmia indigenous chickens were analyzed to identify SNPs associated with birth weight. The objective was to reduce genotyping costs while maintaining prediction accuracy by selecting a subset of highly informative SNPs. After quality control, 45,512 SNPs from 312 F2 chickens were retained for subsequent analyses. SNP importance was quantified using permutation-based variable importance (%IncMSE), and selected marker subsets were evaluated through genomic prediction analyses. The RF algorithm identified informative SNPs with substantial variation in importance among chromosomes. Marker subsets selected using stringent importance thresholds achieved higher genomic prediction accuracy than the complete SNP dataset, indicating that many markers contributed little to prediction performance. Several chromosomes contained few or no highly informative markers, suggesting that they may have limited value for trait-specific low-density SNP panel development. Overall, integrating machine learning-based feature selection with genomic prediction represents a promising approach for developing cost-effective low-density SNP panels to support genomic selection in broiler breeding programs.