Bio Copy Number Hrd Scoring

Version Compatibility

Reference examples tested with: R 4.3+ with scarHRD 0.1.1+, sequenza 3.0+ (allele-specific input); HRDetect / CHORD as their respective R packages where whole-genome data is available.

Before using code patterns, verify installed versions match. If versions differ:

• R: packageVersion('scarHRD') then ?scar_score to confirm arguments
• scarHRD is GitHub-only (sztup/scarHRD); install with remotes::install_github

scarHRD consumes allele-specific copy number — a Sequenza .seqz file or an ASCAT/allele-specific segment table. It cannot run on relative log2 copy ratio.

HRD Scoring

"Is this tumor homologous-recombination deficient" -> HRD leaves characteristic copy-number scars. Three are quantified and summed into an HRD score: loss of heterozygosity (LOH), large-scale state transitions (LST), and telomeric allelic imbalance (TAI). A high score predicts response to platinum chemotherapy and PARP inhibitors. The scar score is a consequence of past HR deficiency — which is both its strength (it integrates over tumor history) and its key limitation.

• R: scarHRD — the three genomic scars and their sum
• Whole-genome: HRDetect (weighted multi-signature model), CHORD (random forest)
• Input: allele-specific copy number from Sequenza or ASCAT (see allele-specific-copy-number)

The Three Genomic Scars

The HRD score is the sum of the three (the "genomic instability score", GIS). Each component has a precise size rule — these thresholds (15 Mb, 10 Mb, 3 Mb) are not arbitrary; they were selected to correlate with BRCA1/BRCA2/RAD51C deficiency (Abkevich 2012, Popova 2012, Birkbak 2012).

Method Selection

Decision: for a targeted panel or WES the genomic-scar score (scarHRD-style) is the only option and is the basis of approved companion diagnostics; for whole-genome data, HRDetect or CHORD are more accurate because they add mutational-signature evidence.

Computing Genomic Scars with scarHRD

Goal: Compute LOH, LST, TAI, and the HRD sum from allele-specific copy number.

Approach: Run scarHRD on a Sequenza .seqz file (or an allele-specific segment table); supply the genome build and ploidy so LST is correctly normalized.

library(scarHRD)

# From a Sequenza .seqz file (allele-specific copy number, with BAF).
hrd <- scar_score('sample.small.seqz.gz',
                  reference = 'grch38',
                  seqz = TRUE)
# hrd is a one-row data frame with columns 'HRD' (LOH), 'Telomeric AI', 'LST', 'HRD-sum'.

# From a pre-computed allele-specific segment table (ASCAT-style: SampleID, Chromosome,
# Start_position, End_position, total_cn, A_cn, B_cn, ploidy):
hrd_seg <- scar_score('sample_allele_specific.txt',
                      reference = 'grch38', seqz = FALSE)
print(hrd_seg)

The Postdoc-Level Caveats

Three points separate a correct HRD interpretation from a naive one:

1. HRD is a scar, not a current state. The score reflects HR deficiency that occurred during tumor evolution. A tumor that has acquired a BRCA reversion mutation — a real platinum/PARP-inhibitor resistance mechanism — still carries the scars and still scores HRD-high. A high score is not a guarantee of current HR deficiency or of drug response.
2. LST is ploidy-dependent. Whole-genome doubling adds breakpoints and inflates the LST count independently of HR status. The score must be computed with the correct ploidy so LST is normalized; an uncorrected WGD tumor can score falsely high.
3. The score needs allele-specific input. LOH and TAI are allelic-imbalance metrics — they cannot be derived from total copy number or relative log2. Garbage allele-specific input (low purity, sparse hets) gives a garbage score.

Failure Modes

Relative copy number used as input

Trigger: Feeding log2 copy ratio or total-CN segments to a scar calculator.

Mechanism: LOH and TAI require the minor allele copy number; relative or total CN has no allelic information.

Symptom: LOH and TAI near zero regardless of true HRD; nonsensical score.

Fix: Use allele-specific copy number from Sequenza or ASCAT (allele-specific-copy-number). The .seqz file or an A/B-allele segment table is the correct input.

LST inflated by uncorrected whole-genome doubling

Trigger: Running the scar score without supplying the tumor's ploidy, on a WGD tumor.

Mechanism: WGD multiplies segments and breakpoints; LST counts breaks and rises with ploidy independent of HR deficiency.

Symptom: A WGD tumor with no BRCA/HR pathway lesion scores HRD-high, driven by LST.

Fix: Compute the score with the correct ploidy so LST is normalized. Cross-check a high LST-driven score against HR-pathway gene status and against mutational signature 3.

Treating a high score as proof of drug response

Trigger: Equating HRD-high with current HR deficiency and predicted PARP-inhibitor benefit.

Mechanism: The scar persists after HR function is restored (BRCA reversion, other resistance mechanisms); the score integrates over history.

Symptom: An HRD-high tumor fails to respond; the score was correct but the tumor is no longer HR-deficient.

Fix: Interpret the score as evidence of past HRD. Where possible, integrate current HR-pathway status (BRCA1/2 reversion screening, RAD51 foci assays) before predicting response.

Low tumor purity

Trigger: Computing HRD on a low-purity sample (< ~30-40%).

Mechanism: Allele-specific calling fails at low purity (see allele-specific-copy-number); scar counts then derive from an unreliable profile.

Symptom: Score unstable across reruns; LOH/TAI near zero on a genome with visible imbalance.

Fix: Confirm purity is adequate before scoring; report indeterminate below ~30%.