Bio Copy Number Focal Amplification Ecdna

Version Compatibility

Reference examples tested with: AmpliconSuite-pipeline 1.3+, AmpliconArchitect 1.3+, AmpliconClassifier 1.2+, CNVkit 0.9.10+, Python 3.10+, samtools 1.19+.

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

• CLI: AmpliconSuite-pipeline.py --help, amplicon_classifier.py --help
• AmpliconArchitect needs a $AA_DATA_REPO reference download and a Mosek license (free for academic use); confirm both are configured before running

Verify the reference build — AmpliconArchitect was historically hg19-centric; GRCh38 support and data repos exist but the build must be set explicitly and consistently.

Focal Amplification and ecDNA

"This oncogene is amplified — but how, structurally" -> A depth caller reports "high focal amplification" and stops. The biology depends entirely on the architecture: extrachromosomal DNA (ecDNA) behaves utterly differently from a chromosomal homogeneously staining region. Resolving architecture needs the breakpoint graph, not depth.

• CLI: AmpliconSuite-pipeline.py (end-to-end), AmpliconArchitect (graph reconstruction), AmpliconClassifier (architecture call)
• Input: WGS BAM plus copy-number seeds (high-CN focal regions)

Why Architecture Matters — Four Amplicon Classes

ecDNA is the highest-stakes call: because it lacks a centromere it segregates unequally, so copy number can surge under selection — a structural basis for resistance. Depth alone cannot distinguish ecDNA from an HSR; both look like a high-amplitude focal gain.

When to Suspect ecDNA

The AmpliconSuite Workflow

AmpliconArchitect does not call amplifications from scratch — it reconstructs the architecture of the regions it is seeded with. The pipeline is: (1) call copy number and select high-CN focal seeds, (2) AmpliconArchitect builds the breakpoint graph and optimizes a balanced flow, (3) AmpliconClassifier labels each amplicon ecDNA / BFB / HSR / linear.

# End-to-end: AmpliconSuite-pipeline runs CNVkit seeding, AmpliconArchitect, and
# AmpliconClassifier in sequence.
AmpliconSuite-pipeline.py \
    -s sample_id \
    -t 8 \
    --bam tumor.bam \
    --ref GRCh38 \
    --run_AA --run_AC

# Output: per-amplicon breakpoint graphs, cycles files, and an AmpliconClassifier
# table assigning each amplicon an architecture class.

Supplying explicit seeds (recommended when a vetted CNV callset exists):

# Seeds: a BED of high-copy focal regions (e.g. from cnvkit-analysis), filtered to
# CN above the seed threshold and to focal (not arm-level) size.
AmpliconSuite-pipeline.py -s sample_id -t 8 --bam tumor.bam --ref GRCh38 \
    --cnv_bed focal_seeds.bed --run_AA --run_AC

Failure Modes

Garbage copy-number seeds produce garbage amplicons

Trigger: Seeding AmpliconArchitect with a noisy CNV callset, a flat-reference tumor-only callset, or arm-level segments.

Mechanism: AmpliconArchitect reconstructs the architecture of exactly the regions it is seeded with; false high-CN seeds generate spurious amplicons, and arm-level seeds dilute the focal signal.

Symptom: Implausible amplicons at no known oncogene; amplicons spanning whole arms; classifier output dominated by low-confidence calls.

Fix: Seed only vetted, focal, high-CN regions. Build the CNV callset from a proper panel of normals (see cnvkit-analysis); filter to focal size and CN above the seed threshold before passing to AA.

Calling ecDNA from depth alone

Trigger: Labeling a high-amplitude focal gain "ecDNA" without breakpoint-graph evidence.

Mechanism: ecDNA and a chromosomal HSR both present as high focal copy number; only the breakpoint graph (a closed cycle with balanced flow) distinguishes them.

Symptom: ecDNA claimed from a CNVkit/GATK profile; no graph, no cycle.

Fix: Require AmpliconArchitect graph reconstruction and an AmpliconClassifier ecDNA call. Where feasible, confirm with orthogonal evidence — FISH, single-cell copy number (variable per-cell CN), or optical mapping.

Genome-build mismatch

Trigger: BAM aligned to one build, --ref or $AA_DATA_REPO set to another.

Mechanism: Coordinates and the bundled annotation diverge; breakpoints and genes are mis-assigned.

Symptom: Amplicons at wrong loci; AA errors on contig names.

Fix: Set --ref to match the BAM's build and confirm the corresponding $AA_DATA_REPO is installed; AA was historically hg19-centric, so GRCh38 must be explicit.

Short-read limits on complex amplicon resolution

Trigger: Expecting a fully resolved amplicon structure from short-read WGS on a highly rearranged amplicon.

Mechanism: Short reads cannot phase long-range structure or traverse repeats; complex amplicons (many junctions, segmental duplications) are only partially reconstructed.

Symptom: Fragmented breakpoint graph; ambiguous or "unknown" classifier calls on a clearly amplified locus.

Fix: Treat short-read amplicon structure as a hypothesis for the most complex cases; confirm with optical mapping (AmpliconReconstructor) or long-read sequencing.