DeepCNV

DeepCNV applies deep learning to assess and improve copy number variation (CNV) calls, increasing the accuracy and validation of duplications and deletions in genomic datasets.

Key Features:

• Deep neural network: Uses a deep neural network algorithm to re-score and validate CNV calls.
• PennCNV integration: Operates on CNV calls initially produced by PennCNV.
• Expert-scored training data: Trained on an extensive dataset of over 10,000 expert-scored samples partitioned into training and testing sets.
• Performance metric: Demonstrates an area under the ROC curve (AUC) of 0.909.
• Wet-lab corroboration: Accuracy was corroborated using experimental wet-lab validation datasets.
• False positive reduction: Produces a significant reduction in false positive CNV calls.
• Reproducibility enhancement: Improves reproducibility of CNV association studies.
• Manual validation replacement: Intended to replace manual expert validation of CNV calls.

Scientific Applications:

• CNV detection and validation: Enhances detection and validation of copy number duplications and deletions in genomic analyses.
• CNV association studies: Increases reproducibility and confidence in studies linking CNVs with disease phenotypes.
• Genetic research and clinical use: Facilitates more precise genetic research and potential clinical applications by improving CNV call reliability.

Methodology:

A deep neural network was trained on over 10,000 expert-scored samples split into training and testing sets to re-evaluate CNV calls produced by PennCNV, with performance assessed by AUC (0.909).