Enigma structural variant implementation
How Enigma Clinical annotates and scores copy-number variants, other structural variants, mobile element insertions and gene fusions
- 80
- rubric criteria (loss + gain)
- 30
- computed automatically
- 5
- proposed for review
- 45
- curator only
Overview
Enigma Clinical shows a page for every structural event it can name on GRCh38: deletions, duplications, inversions, copy-number and complex calls given as an interval, insertions of known length, translocation breakend pairs, mobile element insertions, and gene fusions given as an ordered gene pair with or without breakpoints. Each page lists the genes and transcripts the event touches, ClinGen dosage curations, population records in gnomAD, structural records in ClinVar and publications for the region. Deletions and duplications are additionally scored against the ACMG/ClinGen technical standard for constitutional copy-number variants (Riggs et al. 2020(opens in a new tab)), using the ClinGen dosage map (dosage sensitivity curation(opens in a new tab)) and the same points as the ClinGen CNV calculator(opens in a new tab).
The score is a provisional starting point. Only the parts of the rubric that can be read from public data are counted: genomic content (section 1), overlap with dosage-sensitive and benign regions (section 2), the number of protein-coding genes (section 3) and overlap with common population variation (4O). Case evidence, phenotype fit, inheritance and segregation are scored by a curator on the page itself, and a suggested score is never counted until someone applies it. Nothing a curator enters is stored.
Identifiers and pages
Events are addressed by one canonical identifier each (defined in src/lib/variant-id.ts and src/lib/fusion-id.ts). Coordinates are 1-based GRCh38, chromosomes are written without a “chr” prefix, and search accepts the usual alternative spellings. Translocation breakends are stored with the lower chromosome first; fusion partners keep their biological 5′ → 3′ order.
| Event | Canonical ID | Example | Page |
|---|---|---|---|
| Deletion, duplication, inversion, CNV, complex | {chr}-{start}-{end}-{DEL|DUP|INV|CNV|CPX} | 15-100013402-100497325-DEL | /sv/ |
| Insertion of known length | {chr}-{pos}-L{length}-INS | 2-1234567-L350-INS | /sv/ |
| Translocation / breakend pair | {chrA}-{posA}-{chrB}-{posB}-BND | 9-130714000-22-23290000-BND | /sv/ |
| Mobile element insertion | {chr}-{pos}-{ALU|L1|SVA|HERVK} | 1-1234567-ALU | /mei/ |
| Gene fusion (gene level) | {GENE5}::{GENE3} | BCR::ABL1 | /fusion/ |
Each interval page also derives an ISCN-style name (ISCN 2024(opens in a new tab); for example arr[GRCh38] 15q26.3(100013402_100497325)x1) and the cytogenetic band range from the bundled UCSC hg38 band table.
What every structural event page computes
Genes and overlap
Genes come from the gnomAD API (GENCODE v39 models) for the interval, or for the single position of an insertion or breakend. For each gene the page records whether it is protein-coding (it has coding exons on its canonical transcript), how many of its canonical exons the event hits, and the overlap class: whole (gene inside the event), partial (one breakpoint inside the gene), internal (both breakpoints inside the gene), or breakend / insertion for point events. Coding and non-coding genes are counted and listed separately.
For a gene that is cut, the breakpoint is read on the canonical transcript in the direction of the MANE Select strand: which end (5′ or 3′) the event covers, which exons are hit in transcript order, how many coding bases of the transcript lie inside the event, whether only whole exons are affected (no breakpoint falls inside an exon) and whether the last exon is involved. These values drive the 2C–2E and 2I criteria.
Transcripts
For interval events up to 10 Mb, every transcript in the interval is listed with its MANE flag (Select or Plus Clinical) and RefSeq ID from the bundled MANE v1.5 summary (Morales et al. 2022(opens in a new tab)), whether it is the gene’s canonical transcript in gnomAD, and three overlap shares: genomic span, exonic bases and coding bases inside the event.
ClinGen dosage sensitivity
Haploinsufficiency and triplosensitivity scores for genes and regions come from the ClinGen dosage lists bundled with the platform (scores 0–3, 30 for autosomal recessive, 40 for dosage sensitivity unlikely), with the curated condition. For each overlapping region the page keeps two shares: how much of the ClinGen region the event covers, and how much of the event the region covers. A region counts as wholly covered at 99.9% or more.
Population records (gnomAD)
For intervals up to 10 Mb the page reads gnomAD-SV v4 (Collins et al. 2020(opens in a new tab)) and the gnomAD v4 exome copy-number calls over the interval. Each record carries its reciprocal overlap with the query — the overlapping bases divided by the longer of the two intervals — and whether it is the same class (deletions match DEL and CNV records, duplications match DUP and CNV, inversions and complex events match only themselves). Records are ranked same-type first, then by reciprocal overlap; the count and the highest frequency of same-type records with at least 50% reciprocal overlap are reported. Insertions and mobile element insertions search a ±1 kb window and match by element (ALU, LINE1, SVA, HERVK, read from the record’s ALT) or by length similarity; translocations search ±1 kb around the first breakend and call a record the same type when its partner breakend lies within 1 kb of the second.
ClinVar, publications and phenotype
ClinVar is searched live for structural records of 50 bp or more whose GRCh38 location overlaps the event (either end inside the interval, or spanning it); each record is shown with its classification, review status, conditions, reciprocal overlap or distance, and whether it is the same class (losses with deletions and copy-number losses, gains with duplications and copy-number gains). Publications are gathered from Europe PMC by three kinds of topic: the cytoband in the title with an event word (deletion, duplication, inversion, translocation, mobile element…), the condition of an overlapping ClinGen region, and up to six dosage-relevant genes. A curator may also enter up to eight HPO terms; the page reports which genes in the event are annotated to each term or its descendants.
CNV classification (deletions and duplications)
The rubric follows Riggs et al. 2020(opens in a new tab), Tables 1 (losses) and 2 (gains), in five sections. The CNV page groups criteria by section, shows which were met, not met or proposed for review, lists the genes or regions behind each decision, and lets a curator pick points for the remaining criteria from the rubric’s own ranges. The provisional total is the sum of the counted points; the category thresholds are those of the standard.
- Pathogenic
- 0.99 or more
- Likely pathogenic
- 0.90 to 0.98
- Uncertain significance
- −0.89 to 0.89
- Likely benign
- −0.90 to −0.98
- Benign
- −0.99 or less
Sections
- Section 1
- Genomic content
- Section 2
- Overlap with dosage-sensitive or benign genes and regions
- Section 3
- Number of protein-coding genes
- Section 4
- Literature, databases and case evidence
- Section 5
- Inheritance and family history of the patient
How the automated sections are decided
- Section 1: 1A is met when the event contains a protein-coding gene or a ClinGen-curated region with a score above 0 (other than 40); otherwise 1B (−0.60) is met, with a note when only non-coding genes are present.
- Loss 2A (+1.00): a gene with HI score 3 lies wholly inside the event, or an HI-3 region is wholly covered. 2B is flagged when an HI-3 region is only partly covered.
- Loss 2C–2E: for each HI-3 gene the event cuts, the breakpoint detail chooses 2C-1 (5′ end, coding lost, +0.90), 2C-2 (5′ UTR only, 0), 2D-1 (3′ UTR only, 0), 2D-3 (only the last exon, +0.30; 2D-2 is suggested if pathogenic variants are known there), 2D-4 (3′ end beyond the last exon, +0.90) or 2E (both breakpoints inside the gene). 2E counts PVS1 (+0.90) only for whole-exon, out-of-frame losses that spare the first and last exon, following the PVS1 decision tree (Abou Tayoun et al. 2018); in-frame or exon-cutting cases are proposed for review.
- Loss 2F (−1.00) / 2G (0): a ClinGen region scored 40 contains the whole event, or merely overlaps it.
- Loss 2H (+0.15): when 2A is not met and constraint was looked up (up to 40 protein-coding genes), a gene with gnomAD v4 pLI ≥ 0.9 and LOEUF < 0.6 counts as predicted haploinsufficient (Karczewski et al. 2020; v4 recommends LOEUF < 0.6 where v2 used 0.35).
- Gain 2A (+1.00) / 2B: as for losses, with TS score 3. Gain 2C–2G: against ClinGen regions scored 40, the strongest benign criterion is kept — 2C (−1.00) when the event and the region cover 95% or more of each other, 2D (−1.00) when the region contains the event and no protein-coding gene is cut, 2E (0) when one is, 2F (−0.90) when the event contains the region without adding a protein-coding gene, 2G (0) otherwise.
- Gain 2H–2L: an HI-3 gene wholly inside a gain is noted (2H, 0); an HI-3 gene with both breakpoints inside is proposed as 2I with PVS1_Strong (+0.45) when the gained exons are whole and out of frame; one breakpoint in an HI-3 gene is 2J (0; 2K needs the phenotype); breakpoints in other protein-coding genes are 2L (0).
- Section 3: protein-coding genes wholly or partly in the event — losses 3A 0–24 (0), 3B 25–34 (+0.45), 3C ≥ 35 (+0.90); gains 3A 0–34, 3B 35–49, 3C ≥ 50.
- Section 4O (−1.00): a same-type gnomAD-SV v4 or exome CNV record with at least 50% reciprocal overlap has a frequency of 1% or more. When gnomAD was not compared (interval above 10 Mb or request failed) 4O is marked for review.
- Sections 4A–4N and 5: curator only. ClinVar records of similar content are laid out beside section 4; section 5 is chosen from an inheritance menu that carries the rubric’s own point values (for example confirmed de novo with a highly specific, unique phenotype +0.45; inherited from an unaffected parent with a specific phenotype −0.30).
Copy-number loss rubric
| Code | Section | Criterion | Points | On this platform | How it is evaluated |
|---|---|---|---|---|---|
1A | 1 · Genomic content | Contains protein-coding genes or other known functionally important elements | 0.00 | Automated | Met when a protein-coding gene (GENCODE, coding exons) or a ClinGen-curated region lies in the interval. |
1B | 1 · Genomic content | Contains no protein-coding genes and no known functionally important elements | -0.60 | Automated | Met when neither is present. Non-coding genes alone are flagged for review. |
2A | 2 · Overlap with dosage-sensitive or benign genes and regions | Complete overlap of an established haploinsufficient (HI score 3) gene or region | +1.00 | Automated | Met when a gene with HI score 3 lies wholly inside, or a region with HI score 3 is wholly covered. |
2B | 2 · Overlap with dosage-sensitive or benign genes and regions | Partial overlap of an established HI region (causative gene or critical region not shown to be included) | 0.00 | Proposed for review | Flagged when an HI-3 region is only partly covered; the curator checks its critical genes. |
2C-1 | 2 · Overlap with dosage-sensitive or benign genes and regions | Partial overlap of an HI gene's 5′ end (3′ end not involved), coding sequence included | +0.90 (+0.45 to +1.00) | Automated | Read on the MANE strand and canonical exons: the 5′ end is covered and coding bases are lost. |
Copy-number gain rubric
| Code | Section | Criterion | Points | On this platform | How it is evaluated |
|---|---|---|---|---|---|
1A | 1 · Genomic content | Contains protein-coding genes or other known functionally important elements | 0.00 | Automated | Met when a protein-coding gene (GENCODE, coding exons) or a ClinGen-curated region lies in the interval. |
1B | 1 · Genomic content | Contains no protein-coding genes and no known functionally important elements | -0.60 | Automated | Met when neither is present. Non-coding genes alone are flagged for review. |
2A | 2 · Overlap with dosage-sensitive or benign genes and regions | Complete overlap of an established triplosensitive (TS score 3) gene or minimal critical region | +1.00 | Automated | Met when a gene with TS score 3 lies wholly inside, or a region with TS score 3 is wholly covered. |
2B | 2 · Overlap with dosage-sensitive or benign genes and regions | Partial overlap of an established TS region (causative gene or critical region not shown to be included) | 0.00 | Proposed for review | Flagged when a TS-3 region is only partly covered. |
2C | 2 · Overlap with dosage-sensitive or benign genes and regions | Identical in gene content to an established benign copy-number gain | -1.00 | Automated | Met when the interval and a ClinGen benign gain region (TS 40) cover each other almost entirely (95% or more). |
Inversions, complex events, insertions, translocations and MEIs
These events are not scored. Inversion and complex pages show the same gene, transcript, dosage, gnomAD, ClinVar and publication context as a CNV. Insertion, translocation and mobile element pages show the genes at each site, ClinGen curations at the site, nearby gnomAD-SV records (insertions and mobile elements by element or length, translocations by breakend pairing), ClinVar records within 100 bp (insertions and MEIs) or 1 kb (breakends, limited to translocation, complex and fusion records) and publications for the band and genes. Mobile element insertions accept the four classes gnomAD-SV annotates: Alu, LINE-1, SVA and HERV-K.
Gene fusions
A fusion page describes an ordered 5′::3′ gene pair, optionally with a GRCh38 breakpoint for each partner. Partner genes are read from the gnomAD API (coordinates, strand, canonical and MANE Select transcripts with exons) and shown with their cytoband; a supplied breakpoint is placed on the partner’s transcript track and shaded on the retained side according to strand and 5′/3′ role. Evidence comes from CIViC: the fusion feature whose 5′ and 3′ genes match the pair exactly, its molecular profiles, and the accepted or submitted evidence items (level, type, direction, significance, disease, therapies, source) for profiles that contain the fusion alone; profiles that also require other variants are listed separately and their evidence is not attributed to the fusion. Publications are the ten most cited Europe PMC records naming the fusion, and matching registry trials are listed from the trials search. Exon junction, reading frame, retained protein domains and an oncogenicity call are not computed.
Limits and cut-offs
| Area | Parameter | Value | In the code |
|---|---|---|---|
| Scope | Interval length up to which genes and transcripts are listed | 30 Mb | MAX_GENES_SPAN |
| Scope | Interval length up to which gnomAD records are compared | 10 Mb | MAX_RECORDS_SPAN |
| Scope | Interval length up to which transcripts are listed | 10 Mb | MAX_TRANSCRIPTS_SPAN |
| Scope | Window around insertion sites and breakends | ± 1,000 bp | POINT_WINDOW |
| Population | Reciprocal overlap for a gnomAD record to count as "similar" | ≥ 50% | section4o / counts |
Data sources
Genes, canonical-transcript exons and all transcripts in the interval; gene models at insertion sites and breakends. Used for: Gene and transcript tables, exon and coding-base counts, sections 1, 2 and 3.
Released without restrictionpLI and LOEUF (oe_lof_upper) for the protein-coding genes of a loss. Used for: Loss 2H.
Released without restrictionStructural variants and exome copy-number calls in the window, with allele or site frequencies. Used for: Population overlap tables, 4O.
Released without restrictionHaploinsufficiency and triplosensitivity scores of genes and regions, with the curated condition. Used for: Dosage section, sections 1 and 2, publication topics.
CC0MANE Select and MANE Plus Clinical transcripts, RefSeq IDs and strand. Used for: 5′ / 3′ orientation of partial overlaps, transcript flags.
Public domain (NCBI) / open (EMBL-EBI)Cytogenetic bands. Used for: Band names, ISCN strings, publication searches.
Free data trackStructural records (≥ 50 bp) whose GRCh38 location overlaps the event, with classification, review status and conditions. Used for: ClinVar section; section 4 case review.
Public domain (US government)Titles and metadata of papers matching the cytoband, the ClinGen condition, the dosage-sensitive genes, or the fusion name. Used for: Publications sections.
Open (metadata)Term names and the genes annotated to a term or its descendants. Used for: Optional phenotype match on the CNV page.
HPO licence (free with attribution)Fusion feature, molecular profiles and evidence items for the ordered gene pair. Used for: Fusion evidence.
CC0Sources not used: DECIPHER, DGV (bulk data), COSMIC fusions, OncoKB, Mitelman database, VarSome.
Limitations
- Sections 4 (case evidence) and 5 (inheritance) of the ACMG/ClinGen rubric are never scored automatically: there is no patient, so the provisional total only reflects genomic content, dosage overlap, gene count and population frequency.
- Only deletions and duplications are scored. Inversions, complex events, insertions, translocations and mobile element insertions show genes, dosage, population and ClinVar context without points.
- Events are treated as reported: exact breakpoints, copy number, mosaicism and whether a duplication is in tandem are not known, so 2E and 2I (intragenic events) are proposed for review rather than counted unless the loss removes whole exons out of frame.
- Breakpoint detail is read on the canonical transcript of gnomAD’s gene model, oriented by the MANE Select strand; genes without a MANE transcript are flagged for manual reading.
- Intervals above 10 Mb are not compared with gnomAD or ClinVar, and above 30 Mb no genes are listed; the provisional score then carries a review flag for 4O.
- Fusion pages describe the ordered gene pair. Exon junction, reading frame, retained protein domains and oncogenicity are not computed; breakpoints are only placed on the parent transcript tracks.
References
- Riggs ER, et al. Technical standards for the interpretation and reporting of constitutional copy-number variants: a joint consensus recommendation of the ACMG and ClinGen. Genet Med 2020;22:245–257.(opens in a new tab)
- ClinGen Dosage Sensitivity Curation: gene and region haploinsufficiency / triplosensitivity scores (dosage map).(opens in a new tab)
- ClinGen CNV Pathogenicity Calculator (the same point values as Riggs et al. 2020).(opens in a new tab)
- Abou Tayoun AN, et al. Recommendations for interpreting the loss of function PVS1 ACMG/AMP variant criterion. Hum Mutat 2018;39:1517–1524.(opens in a new tab)
- Collins RL, et al. A structural variation reference for medical and population genetics (gnomAD-SV). Nature 2020;581:444–451.(opens in a new tab)
- Karczewski KJ, et al. The mutational constraint spectrum quantified from variation in 141,456 humans (pLI, LOEUF). Nature 2020;581:434–443.(opens in a new tab)
- Morales J, et al. A joint NCBI and EMBL-EBI transcript set for clinical genomics and research (MANE). Nature 2022;604:310–315.(opens in a new tab)
- ISCN 2024: An International System for Human Cytogenomic Nomenclature. Karger, 2024.(opens in a new tab)