CODEX

CODEX performs normalization and copy number variation calling in whole exome sequencing data to detect and characterize CNVs for genomic analyses.

Key Features:

• Normalization Procedure: An advanced cross-sample normalization model that leverages multiple samples and includes terms to mitigate biases from GC content, exon capture efficiency, amplification variability, and latent systemic artifacts.
• Poisson Latent Factor Model: A count-based Poisson latent factor model that explicitly models exome sequencing read counts and accounts for multiple sources of bias and noise.
• Poisson Likelihood-Based Recursive Segmentation: A recursive segmentation procedure based on Poisson likelihood used to identify CNV breakpoints and segments.
• Cross-Sample Normalization: A normalization strategy that reduces noise by normalizing across samples processed through the same sequencing pipeline rather than relying on matched controls.
• Enhanced Detection of Nested CNVs: Segmentation methods tailored to detect complex CNV structures, including nested configurations.

Scientific Applications:

• Population-scale CNV analysis: Evaluation on HapMap samples from the 1000 Genomes Project for population-level CNV assessment.
• Microarray-based validation: Demonstrated superior accuracy across three microarray-based validation datasets.
• Somatic CNV detection in neuroblastoma: Performance assessment on a cohort of 222 neuroblastoma samples with matched normals, including analysis of a rare somatic CNV in ATRX.
• Disease-associated CNV characterization: Detection and characterization of CNVs relevant to human disease studies using whole exome sequencing data.

Methodology:

Computational steps explicitly include cross-sample normalization with terms for GC content, exon capture efficiency, amplification variability and latent artifacts, modeling read counts with a Poisson latent factor model, and Poisson likelihood-based recursive segmentation for CNV calling.