CNValidatron: Accurate And Efficient Validation of PennCNV Calls Using Computer Vision.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Large rare copy number variants (CNVs) are significant drivers of genetic variation, evolution, and disease risk.
• Genotyping arrays are the most common source for CNV detection in large cohorts.
• Existing methods for CNV calling from array data exhibit high false positive rates and are inefficient for genome-wide analysis.

Purpose of the Study:

• To establish the largest dataset of human-verified CNV calls.
• To develop and validate a machine learning model for automated CNV validation.
• To improve the accuracy and efficiency of CNV detection in large genomic datasets.

Main Methods:

• Assembled a dataset of nearly 60,000 human-verified CNV calls from 22,500 samples across multiple cohorts and arrays.
• Utilized visual validation to assess the accuracy of CNV calls, identifying a high proportion of false positives.
• Trained a convolutional neural network (CNN) using a subset of the validated data for automated CNV validation via machine vision.

Main Results:

• Visual validation revealed that 53.7% of CNV calls were false positives and 9.7% were unclear, with significant variation across datasets and genomic regions.
• The developed CNN model achieved over 90% accuracy in automated CNV validation, comparable to human analysts.
• Cross-validation with genome sequencing data confirmed the high accuracy of the visual validation approach.

Conclusions:

• Visual inspection remains the gold standard for validating CNV calls.
• The developed machine learning model effectively automates CNV validation at scale with high accuracy.
• The CNV validation software is available as an R package, facilitating broader application in genomic research.