Penalized classification for optimal statistical selection of markers from high-throughput genotyping: application in sheep breeds

Abstract

The identification of individuals' breed of origin has several practical applications in livestock and is useful in different biological contexts such as conservation genetics, breeding and authentication of animal products. In this paper, penalized multinomial regression was applied to identify the minimum number of single nucleotide polymorphisms (SNPs) from high-throughput genotyping data for individual assignment to dairy sheep breeds reared in Sicily. The combined use of penalized multinomial regression and stability selection reduced the number of SNPs required to 48. A final validation step on an independent population was carried out obtaining 100% correctly classified individuals. The results using independent analysis, such as admixture, F st , principal component analysis and random forest, confirmed the ability of these methods in selecting distinctive markers. The identified SNPs may constitute a starting point for the development of a SNP based identification test as a tool for breed assignment and traceability of animal products.

Genome-wide selection of discriminant SNP markers for breed assignment in indigenous sheep breeds

“…It means that the majority of the SNPs are in non-coding/intergenic regions of the sheep genome which is ideal for identification and assignment purpose since these regions/SNPs should be less influenced by natural or artificial selection (Allen et al, 2010). This result was in agreement with previously published manuscript (Sottile et al, 2018).…”

“…It should be considered that the number of samples may affect the discriminant SNP marker list identified in this study, however, most of the similar published reports have been performed with the same or lower sample size in different indigenous sheep breeds (eg. Dimauro et al, 2015;Tortereau et al, 2017;Sottile et al, 2018). Also, as showed in Supplementary Table 2, a small proportion of the discriminant SNPs identified in current study, are located near the coding genes (56 SNPs out of all discriminant SNPs).…”

Recommendations for Population and Individual Diagnostic SNP Selection in Non-Model Species

“…In addition to estimating individual marker informativeness, there are methods for determining which combinations of markers will yield the most effective panel, such as fORCA (Rosenberg, 2005). This approach has rarely been implemented outside of human and agriculturally-relevant species, such as salmon (Storer et al, 2012), sheep (Sottile et al, 2018), and crop species (Morrell & Clegg, 2007). In rare cases, the method has been applied to nonmodel systems, such as in the domestic cat and European wildcat (Oliveira et al, 2015).…”

Classification of cattle breeds based on the random forest approach

“…The three above-mentioned methods (F ST , PCA and RF) have already been applied to the selection of the most informative SNPs in several animal species such as cattle (Barbato et al, 2020;Bertolini et al, 2015Bertolini et al, , 2018Hulsegge et al, 2013Hulsegge et al, , 2019Wilkinson et al, 2011), sheep (Mastrangelo et al, 2014;Somenzi et al, 2020;Sottile et al, 2018), pig (Chinchilla-Vargas et al, 2021;Schiavo et al, 2020;Zhang et al, 2018), and chicken (Seo et al, 2021;Zhang et al, 2019). However, there are few studies on the application of RF and different variable importance criteria to the identification of the most informative SNPs in dairy and beef cattle breeds and comparison of its predictive performance with more traditional approaches such as the F ST and PCA.…”