Comparison of classification methods for detecting associations between SNPs and chick mortality

Area of Science:

• Animal Genetics
• Bioinformatics
• Poultry Science

Background:

• Chick mortality poses significant economic challenges in the poultry industry.
• Identifying genetic factors influencing mortality is key to improving broiler health and productivity.
• Whole genome single nucleotide polymorphisms (SNPs) offer a high-density marker system for genetic association studies.

Purpose of the Study:

• To select a subset of whole genome SNPs associated with broiler chick mortality.
• To evaluate the effectiveness of multi-category classification methods for SNP selection.
• To compare different categorization schemes and classification algorithms for identifying influential SNPs.

Main Methods:

• Employed multi-category classification and filter-wrapper feature selection to identify SNPs associated with chick mortality.
• Compared naïve Bayes classifiers, Bayesian networks, and neural networks across 2, 3, 4, 5, and 10 mortality categories.
• Utilized predicted residual sum of squares and significance test metrics for SNP evaluation.
• Investigated an alternative categorization using only extreme mortality rates.

Main Results:

• A naïve Bayes classifier with 2 or 3 mortality categories yielded the most predictive SNP subset.
• An alternative categorization scheme focusing on extreme mortality rates demonstrated superior SNP predictive ability.
• The use of extreme samples enhanced the selection of influential SNPs in genetic association analyses.

Conclusions:

• Feature selection methods, particularly those using extreme sample categorization, are effective for identifying SNPs related to broiler chick mortality.
• Naïve Bayes classifiers combined with appropriate data discretization show promise for genetic association studies in poultry.
• Optimizing SNP selection strategies can improve the accuracy of identifying genetic markers for economically important traits like disease resistance.