An Explainable Deep Learning Classifier of Bovine Mastitis Based on Whole-Genome Sequence Data-Circumventing the p >>...

Area of Science:

• Genomics
• Veterinary Medicine
• Bioinformatics

Background:

• The p >> n problem (many polymorphic variants, few phenotypic records) hinders biological annotation of whole-genome data.
• Accurate prediction of disease susceptibility in livestock is crucial for herd management and breeding programs.

Purpose of the Study:

• To develop and validate a machine learning model for classifying cows as susceptible or resistant to mastitis.
• To address the challenge of high-dimensional genomic data in biological annotation.

Main Methods:

• A hybrid approach combining LASSO logistic regression with deep learning was employed.
• A deep learning architecture with 204,642 single nucleotide polymorphisms (SNPs) and specific layer configurations was optimized.
• SHapley Additive exPlanation (SHAP) was used to identify significant SNPs.

Main Results:

• The best deep learning model achieved an Area Under the Curve (AUC) of 0.750, accuracy of 0.650, sensitivity of 0.600, and specificity of 0.700.
• Significant SNPs were identified, and enriched Gene Ontology (GO) terms related to immune response and protein synthesis were found.
• The model correctly predicted susceptibility or resistance status for approximately 65% of cows.

Conclusions:

• The combined LASSO logistic regression and deep learning model effectively predicts mastitis susceptibility in cows.
• The identified significant SNPs and associated genes provide insights into the genetic basis of mastitis resistance.
• This approach offers a viable solution for the p >> n problem in genomic data analysis for cattle.