Visualization of Runs of Homozygosity and Classification Using Convolutional Neural Networks

Area of Science:

• Genetics
• Bioinformatics
• Machine Learning

Background:

• Runs of homozygosity (ROH) are crucial for understanding genetic structure, inbreeding, and selection history in populations.
• ROH analysis traditionally requires complex computational methods, limiting its widespread application.

Purpose of the Study:

• To develop and validate a novel approach for ROH analysis using convolutional neural networks (CNNs).
• To classify pig breeds and identify phenotypic traits based on visualized ROH maps.
• To explore the application of this method in animal breeding and human medicine.

Main Methods:

• Genetic data from Large White and Duroc pigs were used to create ROH maps, visualizing homozygous segments.
• Convolutional neural networks (CNNs) were employed for classification tasks: breed identification and limb defect prediction.
• ROH segments were identified using PLINK v1.9, and visualization was achieved with a modified HandyCNV package function.

Main Results:

• The CNN model achieved 100% accuracy, sensitivity, and specificity in classifying pig breeds based on ROH maps.
• The model showed 78.57% accuracy in predicting the presence or absence of limb defects.
• A high negative predictive value (84.62%) was observed for identifying healthy animals regarding limb defects.

Conclusions:

• CNN-based ROH map visualization offers a highly accurate method for breed classification in pigs.
• The approach shows potential for identifying genetic predispositions to phenotypic traits like limb defects, with implications for disease association studies.
• This innovative method can be extended to diverse genetic data and applied in both animal breeding and medical research for improved genetic insights.