DECoNT

DECoNT refines copy number variant (CNV) calls from whole exome sequencing (WES) data to increase the precision of duplication and deletion detection for genetic research and clinical diagnostics.

Key Features:

• Deep Learning-Based Correction: Employs a deep learning model trained to refine CNV predictions derived from WES datasets.
• Use of Matched WGS Data: Leverages matched whole genome sequencing (WGS) data alongside WES to learn and correct biases inherent in exome-based CNV detection.
• Universal Polisher: Functions as a universal polisher for CNV calls across different sequencing technologies, exome capture kits, and CNV callers.

Scientific Applications:

• Improved Precision: Correcting CNV predictions increases precision of duplication and deletion calls, reported to triple duplication precision and double deletion precision compared to state-of-the-art algorithms.
• WES-based CNV Studies: Enables more accurate germline CNV detection from cost-effective WES data, supporting large-scale genetic studies and clinical analyses.

Methodology:

DECoNT uses a deep learning model trained on 1000 Genomes Project data with matched WGS–WES samples to correct exome-derived CNV predictions and mitigate biases from non-contiguous targeted capture, genomic hybridization effects, GC-content variation, probe targeting issues, and sample batching during sequencing.