Classifying aneuploidy in genotype intensity data using deep learning.

Area of Science:

• Genetics
• Bioinformatics
• Animal Breeding

Background:

• Aneuploidy, the gain or loss of chromosomes, is rare in live births but occurs in livestock breeding.
• Current aneuploidy detection relies on time-consuming visual inspection of SNP array intensity data.
• Automated methods are needed to efficiently screen large breeding populations.

Purpose of the Study:

• To develop and validate a deep learning Convolutional Neural Network (CNN) model for automated aneuploidy classification.
• To classify chromosome-level plots of SNP array intensity data into disomic, monosomic, and trisomic categories.
• To assess the feasibility of automated image classification for routine screening in diploid species.

Main Methods:

• Collected a multispecies dataset with genotype intensity data from 3321 disomic, 1759 monosomic, and 164 trisomic chromosomes.
• Developed a supervised CNN model using chromosome-level plots of SNP array intensity data.
• Compared CNN performance against unsupervised k-means clustering.

Main Results:

• The CNN model achieved 99.9% accuracy, 1.0 precision, 0.98 recall, and 0.99 F1 score.
• CNN significantly outperformed k-means clustering (73% accuracy), especially for trisomic cases.
• High precision indicates reliable detection, though some trisomies might be missed (recall 0.94).

Conclusions:

• A deep learning CNN model accurately classifies aneuploidy from SNP array intensity plots.
• This automated tool can enhance routine screening for genetic anomalies in livestock populations.
• Improved detection aids understanding of aneuploidy incidence, inheritance, and breeding candidate selection.