DeNovoCNN

DeNovoCNN applies a deep convolutional neural network to identify de novo mutations (DNMs) from whole-exome (WES) and whole-genome (WGS) sequencing data for rare disease research and diagnostics.

Key Features:

• Deep Convolutional Neural Network: Employs a CNN architecture to process trio-based genomic sequence alignments for DNM detection.
• Image-like Representation: Transforms alignments into image-like representations at 160×164 resolution for pattern recognition.
• Performance: Achieves 96.74% recall and 96.55% precision on test datasets derived from 5,616 WES trios, outperforming GATK, DeNovoGear, DeepTrio, and Samtools.
• Cross-platform Validation: Validated across sequencing technologies including Sanger and PacBio HiFi sequencing and across different exome sequencing and analysis approaches.
• Input Formats: Operates on alignment files (BAM/CRAM) and variant calling files (VCF) without requiring variant recall.

Scientific Applications:

• Rare Disease Research: Enables accurate identification of pathogenic DNMs in WES and WGS cohorts for rare disease studies.
• Genetic Diagnostics and Disorder Research: Supports diagnostic interpretation and investigation of genetic disorders through reliable DNM detection.

Methodology:

Converts trio-based genomic alignments into 160×164 image-like representations and classifies them with a deep CNN, with evaluation on 5,616 WES trios and validation against Sanger and PacBio HiFi sequencing.