Coverage Depth Normalization

The most common source of non-uniform coverage in sequencing data. Regions with extreme guanine-cytosine (GC) content—either very high or very low—amplify inefficiently during PCR, creating artificial drops in read depth.

• Mechanism: DNA polymerase exhibits sequence-dependent efficiency during amplification
• Correction method: LOESS regression modeling read depth as a function of GC percentage in sliding windows
• Impact: Without correction, GC-rich promoters and GC-poor regulatory deserts appear to have false copy number alterations
• Tools: GATK's CollectGcBiasMetrics, Picard, and deepTools provide GC-normalized coverage tracks

Genomic regions with low sequence complexity or high similarity to other loci produce ambiguous read alignments, artificially reducing apparent coverage depth.

• Blacklisted regions: Centromeres, telomeres, and segmental duplications where reads cannot be uniquely placed
• Mappability score: A continuous metric (0-1) indicating the probability a read of given length maps uniquely to a position
• ENCODE blacklists: Curated sets of problematic regions used as standard filters in production pipelines
• K-mer uniqueness: The foundation of mappability calculation—shorter k-mers have lower uniqueness, explaining why short-read technologies exacerbate this bias

In whole exome sequencing (WES), hybridization probes have variable binding affinities across target regions, creating systematic coverage biases independent of GC content.

• Probe thermodynamics: Melting temperature and secondary structure of baits determine capture efficiency
• Target-specific normalization: Requires per-exon correction factors learned from a panel of normals
• Off-target reads: Reads captured from non-exonic regions dilute effective coverage of targets
• Clinical relevance: Uneven coverage in disease-relevant genes can cause missed diagnoses if not normalized against a matched reference cohort

Coverage depth normalization is the computational foundation for copy number variation (CNV) detection from sequencing data. The normalized read depth at a locus is directly proportional to the underlying DNA copy number.

• Segmentation algorithms: Circular binary segmentation and hidden Markov models partition the genome into regions of constant copy number based on normalized depth
• Panel of normals: A reference set of diploid samples processed identically to establish expected coverage at each genomic bin
• GC and mappability correction: Both must be applied before CNV inference, or systematic biases masquerade as amplifications or deletions
• Single-cell applications: Coverage normalization is especially critical in sparse single-cell data where dropout events are common

After depth normalization, the ratio of alternate to reference alleles at heterozygous sites should approximate 0.5. Systematic deviation from this expectation indicates residual coverage bias or mapping artifacts.

• Expected distribution: Heterozygous germline variants should cluster around 50% allele fraction in diploid regions
• Strand bias detection: Normalized depth per strand reveals whether variants are supported by both forward and reverse reads
• Variant quality recalibration: Tools like GATK VQSR use allele balance as a key annotation feature for distinguishing true variants from artifacts
• Somatic applications: Tumor purity and subclonal architecture are inferred from deviations in allele balance after depth normalization

In whole genome sequencing of proliferating cells, early-replicating regions have higher effective coverage than late-replicating regions because they are present in more copies during S-phase.

• Cell cycle effect: Actively dividing cell populations amplify early-replicating euchromatin relative to late-replicating heterochromatin
• Wavelet smoothing: Used to model and remove the broad, megabase-scale coverage undulations caused by replication timing
• Interaction with GC bias: Replication timing correlates with GC content, requiring joint modeling to avoid overcorrection
• Clinical samples: Tumor biopsies with high proliferation rates exhibit stronger replication timing artifacts than normal tissue