SKYPE is an assembly-based structural-variant caller that uses raw-read depth and junction evidence to support its calls. It builds candidate chromosome paths from assembly junctions, fits their depth contributions, and reports variants with normalized support. Native runs can also rescue missing junction candidates from raw reads in regions with unexplained depth changes.
SKYPE also supports depth annotation of an existing VCF and a separate workflow for complete genome assemblies.
Usage and options · Methods · Output formats and weight interpretation
Run SKYPE through ACCtools-pipeline, which prepares assemblies, alignments, depth and reference resources. First set up the environment in its installation guide. Then, from the ACCtools-pipeline checkout:
mamba activate skype
# PacBio HiFi reads; options precede the working directory.
python SKYPE.py run_hifi --reference hg38 -p SAMPLE -t 16 \
/path/to/work /path/to/reads.fastq.gz
# Reuse an assembly and a PanDepth result.
python SKYPE.py analysis --reference hg38 -p SAMPLE -t 16 \
/path/to/work /path/to/contigs.fa /path/to/unitigs.fa \
/path/to/depth.win.stat.gzUse one reference build for the assembly alignments, depth and any input VCF.
The reference choices are hs1 and hg38; the default is hs1.
See usage for input requirements, ONT/Flye commands, VCF input,
raw-read rescue options and partial restarts. The research workspace's run.sh
is a separate convenience wrapper and is not included in this repository.
| Workflow | Input and purpose | Main result |
|---|---|---|
| Native assembly | Discover NCloses from assembly alignments, fit depth and optionally rescue candidates from raw reads. | SV_call_result.vcf |
| VCF input | Evaluate the supplied variants using SKYPE depth support; select with --benchmark_vcf_loc. |
SV_benchmark_result.vcf |
| Full assembly | Use each supplied FASTA record as a matrix path; select with --full_assembly. |
Separate coverage, Virtual SKY and karyotype outputs |
Native assembly runs default to raw-read rescue (read); VCF input defaults to
rescue off. Local OLC rescue is an explicit option. Full-assembly input uses
its own workflow. See input workflows.
Depth normalization keeps an existing output intact until the replacement gzip is complete, so overlapping runs do not expose a partially written table at the final output path.
An NClose is SKYPE's tracked assembly-junction unit. It includes Type 1,
Type 2 and Type 4 events. An NClose can retain several internal alignment
pieces; it is not necessarily a decomposition into primitive junctions.
The native VCF alone expands each eligible NClose into BNDs between adjacent
alignments. CEN-SAT, telomere-derived and virtual-inversion NCloses, as well as
read and OLC rescue, are included. Normal reference continuations are omitted.
The depth model, BED and plots retain original NCloses.
Native accounting reuses an existing NClose ID when another source pair has
exactly the same two BND boundaries and retained sides. Source pairs and unitigs
remain traceable in nclose_sources.tsv; nearby coordinates are not merged.
A structure carries a weight and a count of how often it uses each original NClose. Structures include chromosome paths, independent Type 4 features, MERGE_TYPE4, AMP and VIRTUAL_INV. Several structures may use the same NClose. See the definitions and pipeline.
Files are written to the selected SKYPE output directory.
| File | What to use it for |
|---|---|
SV_call_result.vcf |
Native variant calls: adjacent-alignment BNDs within eligible constituent NCloses, and symbolic DEL/DUP for ordinary Type 4. |
SV_call_result.nclose_bnds.tsv |
Each original NClose source's junctions and their exact merged VCF BND IDs. |
SV_call_result.bnd_weights.tsv |
Each structure/source NClose's contribution to an exported BND. |
SKYPE_result.bed |
NClose and structure summary views, with their weight scopes identified. |
nclose_report.tsv |
NClose totals after exact BND identity reuse and representative preprocessing history. |
nclose_sources.tsv |
Original node pairs/unitigs and their shared NClose identities. |
structure_report.tsv |
Each native structure's own weight and provenance. |
structure_nclose_usage.tsv |
NClose occurrence counts and each structure's contribution. |
total_cov.png, total_cov.pdf |
Observed and fitted depth with NClose and structure links. |
virtual_sky.png, virtual_sky.pdf |
Virtual SKY diagram of fitted chromosome paths, colored by chromosome of origin and labeled with normalized copy number. |
karyotype.txt |
Tab-separated ISCN-style karyotype labels and their normalized depth (N). |
SV_benchmark_result.vcf |
VCF-input records annotated with SKYPE_CN and SKYPE_STATUS. |
depth_scale.json |
Stage-00 GC correction and the depth per copy (1N) used for the copy numbers above. |
Karyotype mode (the default) also refits the matrix after greedy NClose
filtering and after karyotype clustering, and writes _filter and _cluster
versions of the Virtual SKY, karyotype, Circos, VCF, BED and native accounting
reports. Variant mode (--variant_mode), VCF input and graphs with more than 1000 NClose nodes
write only the unsuffixed results. See
karyotype mode.
The native structure reports do not replace the VCF-input workflow's existing reporting rules. Detailed schemas, IDs and intermediate files are documented in outputs.
Native outputs share this calculation:
NClose CN = sum(structure weight × NClose occurrence count) / N
BND CN = sum(structure weight × source NClose occurrence count × BND occurrences in that source) / N
N = depth per copy (1N) recorded in depth_scale.json
Stage 00 estimates N from the sample's B-allele fractions (allelic states such as
1:1, 1:2 and loss of heterozygosity must sit on whole copy numbers), because the median
depth is not two copies in a near-triploid or near-tetraploid sample. Without that
estimate, N is the median read depth excluding chrM / 2. Every cut that only asks
whether a value is small, such as CN > 0.1 and the 0.2N display filters, keeps the
median read depth / 2 as its unit, so a different N changes the reported copy numbers
but not which events are shown.
For example, a path with normalized weight 3 and a virtual inversion with
normalized weight 2 each using NClose A once give A a total of 5. A virtual
inversion contributes its weight to each constituent NClose according to its
usage count; the weight is not divided between its two NCloses.
Likewise, splitting a source NClose gives each distinct junction its full source
contribution. Junctions merge only when both chromosomes, boundaries and retained
sides match exactly, including reverse-complement descriptions. Source-specific
counts keep different internal chains separate even if their outer endpoints match.
These values are support estimates, not read VAFs. Structure CN, NClose CN and
the predicted depth of a reference segment describe different quantities.
Contributions are summed before the native output threshold (CN > 0.1 in median
depth / 2 units).
VCF geometry merging and reciprocal BND mate records can make its
record count differ from the original NClose count.
See weight accounting for AMP normalization, virtual contributions, Type 4 path contributions and worked examples, and output representations for original NClose endpoints and compound summaries.