Detection of 1.4 Mb duplication

[wolfgang-stark]

Can parascopy be used to detect larger duplications e.g. a 1.4 Mb in chromosome 17p11.2 based?
If so, would 30x fq hg38 data like in this example work?

@E100046761L1C001R00300000033/1
GGACCTAFTTATAGGGTGAGATCTAGGCCTACCTCTGTGATCTCACCCCCCAAACCTCTTCCTGCTGTATTTTCTTTGCAGCACTTATCACTATCTGGAACTATTTACTTTTCTTGTTTATTGTCTGATCTACCACCATTACCGACCCTC
+
GGFGGGGGGGGGGGGGGGGGFGGGGGGGGGGGGGGGGGGFGGGGGGGFGGGGGFGGGGGGGGGGGGGGFGGGGGGGGGGCFGGFGGGGGGGGFGGGFDCEGGGGGGGGGGGGFGGGGGGCGGGGGGG@GGGCGFFGGFGG:FF@;GEGGG
@E100046761L1C001R00300000044/1
TTTCCTGAGGCAATCCCCAGCCATGTGGAACTGTTAGTAACTTAAACTTCTTTCCTTAAGAAATTATCCAGTCTCAGGAAGTGTGAGAATGGACTAATAAAGTAAATTGGTACTGGTAGAGTGAGTACTGCCACAGAGATAGCCAAAAAC
+
ACEC>ED<A2<FF7CBDBC2?EEC>+@EDAF>EFE>FADDFFADCCFFCDFEBAFFEED@EBEFDE::DBF@F@D?.B?<FCD<6.@6CF85AFB@FD@;;DE>=BB=<E<2F+9@C>>E=@67<@D<F;<2E?A;D71DBA77D/?<D=

Thank you very much!

[tprodanov]

Hi @wolfgang-braun , Thank you for using our tool!

Theoretically, it should not be a problem, but you should try and check what happens. If there are any problems, please write us!

[wolfgang-stark]

Hi @tprodanov, coming back to close the loop - it worked 😊! Thank you for this great tool.

Setup

• Sample: 30× WGS, GRCh38 (GCA_000001405.15_GRCh38_no_alt_plus_hs38d1), CRAM input
• Target: CMT1A - the classical ~1.4 Mb duplication on chr17p12 (PMP22 region, NAHR-mediated between CMT1A-REP repeats)
• parascopy version: 1.19.0 (Docker, linux/amd64)
• Homology table: v1.7 GRCh38
• Model: 1KGP EUR pretrained

Results

Tested on two samples (one known CMT1A-positive, one negative control confirmed by clinical Array-CGH & WES):

• Positive sample: parascopy reported psCN = 3,2 across the TVP23C-CDRT4 subregion - correctly identifying the duplication (CN=3 on the chr17p12 paralog, CN=2 on the distal paralog). Independently confirmed by samtools depth ratio (PMP22
  region mean ÷ genome median = 1.50 ≈ CN=3) and a 10 kb-window boundary map placing the duplication at the classical ~1.39 Mb size.
• Negative control: psCN = 2,2 - clean negative, as expected.

Notes for future users trying CMT1A detection

1. parascopy is the right tool for this. Standard CNV callers (Manta, CNVnator, GATK gCNV) struggle with the PMP22 region because it sits within segmentally duplicated sequence flanked by near-identical CMT1A-REP repeats. parascopy's
   paralog-aware model handles this correctly.
2. 30× WGS is sufficient. The call was clean and unambiguous.
3. CRAM input works directly - no need to convert to BAM first.
4. Docker on Apple Silicon (M-series Mac): works fine under --platform linux/amd64 (Rosetta emulation). The PMP22 region is small enough that emulation overhead is negligible.
5. Orthogonal validation is easy: a simple samtools depth ratio (PMP22 region mean ÷ genome median) gives ~1.5 for a duplication carrier, confirming parascopy's call independently.

Thanks again for building and maintaining parascopy - it fills a gap that no other open-source tool covers well for segmentally duplicated CNVs.