CNV-Finder: Streamlining Copy Number Variation Discovery.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Copy Number Variations (CNVs) are crucial in complex disease etiology and population variability.
• Accurate CNV detection is vital for disease genetics research, often requiring large sample sizes.
• Illumina genotyping arrays offer cost-effective CNV detection via Log R Ratio (LRR) and B Allele Frequency (BAF).

Purpose of the Study:

• To develop and validate CNV-Finder, a novel deep learning pipeline for large-scale CNV identification.
• To expedite the prioritization of samples for subsequent whole genome sequencing analyses.
• To enhance the accuracy and efficiency of CNV detection in genes associated with neurological diseases.

Main Methods:

• Integration of deep learning, specifically Long Short-Term Memory (LSTM) networks, with array data (LRR and BAF).
• Training models on expert-annotated samples and validation across diverse cohorts (e.g., Global Parkinson's Genetics Program).
• Development of an interactive web application for visualization, review, and filtering of CNV predictions.

Main Results:

• CNV-Finder accurately detects deletions and duplications, demonstrating efficacy across diverse cohorts.
• The pipeline successfully identifies CNVs in regions with varied sparsity, noise, and size, including complex regions like 17q21.31.
• Human feedback integration enhances model performance and reduces false positive rates.

Conclusions:

• CNV-Finder is a scalable, publicly available resource for efficient and accurate CNV identification.
• The pipeline significantly reduces manual workload for researchers, facilitating targeted validation and downstream analyses.
• Contextual understanding and human expertise are key to enhancing CNV identification precision.