From a variant to a call: Recommended Track Sets, regulation, and the evidence
+
Worked examples in germline & somatic interpretation · genome.ucsc.edu
+
+
+
+
A thread for today: three cancer variants
+
+
A BRCA2 variant: germline, “uncertain significance”. How do experts resolve it? (morning, in the Recommended Track Sets)
+
The TERT promoter: a non-coding driver. The answer lives in the regulation. (morning, then the epigenetics section)
+
BRAF V600E: a coding driver in melanoma. Somatic, famous, druggable. (afternoon, the somatic worked example)
+
+
Watch them recur
+ We will come back to these variants throughout the session. The goal is that by the end you can interpret a variant, explore its regulatory context, load your own data, and share it as a link.
+
+
+
+
+
Interpreting a variant = asking questions (germline)
+
Databases: ask a question, know which track answers it:
+
+
Is it already classified? → ClinVar, ClinGen(germline pathogenicity, ClinVar also carries somatic oncogenicity)
+
Is it druggable? → CIViC (variant → disease → therapy → evidence)
Germline first
+ These questions fit a germline variant, and the Recommended Track Sets (next) bundle exactly these tracks. Later we revisit the same questions for somatic cancer variants.
+
+
+
+
+
Recommended Track Sets
+
Hundreds of tracks is overwhelming, start from a curated set
+
+
+
+
Recommended Track Sets: The problem they solve
+
+
+
+
The Browser has hundreds of tracks, beginners don’t know which to turn on.
+
Recommended Track Sets = pre-configured collections for a scenario.
+
One click turns on a themed set, without changing your locus.
+
Open via the “Recommended Track Sets” menu item.
+
+
+
+ The Recommended Track Sets menu: each link loads a curated, themed set of tracks at your current position.
+
+
+
+
+
+
Seven sets on hg38
+
+
+
+
Clinical SNVs: disease contribution of coding SNVs
+
Clinical CNVs: coding structural variants
+
Non-coding SNVs: functional context of non-coding variants
+
Determine Exon Relevance: is the variant in a required exon?
+
Problematic Regions: low-confidence / high-homology regions
Same codon / nearby seen before? (PS1 / PM5) → ClinVar
+
In a key protein domain? (PM1) → UniProt: the DNA-binding domain (OB1 fold) ✓
+
Computational? (PP3) → REVEL ≈ 0.93 + deep conservation ✓
+
Loss-of-function? (PVS1) → gene model (missense → N/A)
+
+
The point
+ Each line of evidence stacks in one view, here they all agree on pathogenic. You read the ACMG codes off the screen. (The harder, guideline-dependent cases come in the ENIGMA set.)
+
+
+ At p.Asp2723His: RefSeq/MANE, the BRCA2 OB1 UniProt domain (PM1), and ClinVar variants stacked.
+
+
+
+
+
+
Demo 2: Non-coding SNVs → epigenetics (1/2)
+
+
+
For variants outside coding exons, the regulatory evidence:
+
+
GeneHancer enhancers & enhancer→gene links
+
Hi-C / Micro-C 3D chromatin contacts
+
JASPAR TF binding sites · 100-way conservation
+
+
Our bridge to epigenetics later in this deck, and the set for our TERT promoter variant.
+
+
+ Non-coding SNVs near the TERT promoter: GeneHancer regulatory elements, JASPAR TF sites, and conservation, the context a non-coding variant needs.
+
+
+
+
+
+
Demo 2 · cont: a non-coding variant at TERT (2/2)
+
+
+
TERT promoter hotspot mutations (e.g. C228T · NM_198253.3(TERT):c.-124C>T) sit ~100–150 bp upstream of the start codon, in the core promoter. With the Non-coding set, ask:
+
+
In a regulatory element? → GeneHancer / ENCODE cCRE (promoter)
+
Creates / breaks a TF site? → JASPAR (TF motifs here; the C228T / C250T hotspots create a new ETS / GABPA site that switches TERT back on)
+
Contacts a distal gene in 3D? → Hi-C / Micro-C (most useful for enhancer variants; less so for this promoter)
+
Evolutionarily constrained? → conservation
+
+
No coding ACMG here
+ Non-coding variants aren’t scored by coding ACMG rules, you weigh the regulatory evidence instead.
+ The TERT promoter: ENCODE cCRE, GeneHancer, and dense JASPAR TF-binding sites, where the hotspot builds a new ETS site.
+
+
+
+
+
+
Demo 3: expert-panel gene sets
+
+
+
This is the home of our germline BRCA variant of uncertain significance.
+
+
ENIGMA BRCA1/BRCA2 VCEP: the exact evidence for ClinGen ENIGMA classification, per exon & variant.
+
InSiGHT Lynch Syndrome VCEP: same idea for MLH1/MSH2/MSH6/PMS2.
+
+
Same variant, different rules, ▶ open it
+ BRCA2 c.830A>G (p.Asn277Ser): standard ACMG reached likely benign (REVEL → BP4); under ENIGMA it reverts to VUS (BayesDel not allowed here; SpliceAI → PP3). The spec can move a call either way; overall it cut VUS, but not for every variant.
+
Published co-authored
+ Benet-Pagès, Laner, Nassar… Genet Med Open 2025. Session (hg19): /s/abenet/BRCA.ENIGMA.hg19
+
+
+ The ENIGMA BRCA1/BRCA2 VCEP set at a BRCA1 exon (hg19): the exact evidence the panel rules use, per variant.
+
+
+
+
+
+
Somatic diagnosis
+
Somatic variants
+
acquired mutations: SNVs, de novo changes, and cancer drivers
+
+
+
+
Interpreting a variant = asking questions (somatic)
+
Databases: ask a question, know which track answers it:
+
+
Is it already classified? → ClinVar, ClinGen(germline pathogenicity, ClinVar also carries somatic oncogenicity)
+
Is it druggable? → CIViC (variant → disease → therapy → evidence)
Germline vs somatic
+ These questions fit a germline variant. For a somatic driver (BRAF V600E, next) lean on COSMIC & CIViC and ClinVar’s somatic oncogenicity / clinical-impact, and don’t read gnomAD frequency as “benign” (a true somatic variant is just absent). REVEL / conservation flag a damaging residue, not oncogenicity.
+
+
+
+
+
Worked example: BRAF V600E
+
The classic melanoma driver: a somatic variant. BRAF V600Ep.Val600GluNM_004333.6:c.1799T>Achr7:140,753,336
+
+
+
Try it, ▶ open the BRAF V600E session
+ COSMIC: how often in tumours? · CIViC: oncogenic & druggable? · ClinVar: its somatic oncogenicity / clinical-impact, not the germline label.
+
Somatic ≠ germline
+ gnomAD won’t “verify” it: a true somatic variant is simply absent from healthy-population data (gnomAD filters out inherited variants).
+
Thread tie-in
+ Our coding driver: the Recommended “Clinical SNVs” set assembles these in one click.
+
+
+ The BRAF V600E session: ClinVar, COSMIC and CIViC stacked at chr7:140,753,336.
+
+
+
+
+
+
Expression
+
where, and in which cell type, is a gene switched on?
+
+
+
+
Three expression datasets on hg38
+
“Where is my gene expressed, and in which cell type?” Two axes separate these tracks: bulk tissue vs single cell, and one uniform study vs many pooled together.
+
+
+
GTEx Gene V8
+
Bulk tissue reference.
+
+
54 tissues, 948 donors, bulk RNA-seq. On by default.
+
Each bar is a whole tissue, so every cell type is averaged together.
+
Best for: which organ is the gene expressed in?
+
+
+
+
Tabula Sapiens
+
One uniform single-cell atlas.
+
+
~480k cells across ~24 organs, one consortium, consistent processing.
+
Bars split by tissue and by cell type, so signal resolves to a cell type.
+
Best for: which cell type, answered cleanly within one atlas.
+
+
+
+
Merged Single-Cell
+
Many single-cell studies at once.
+
+
Pools many published atlases (incl. Tabula Sapiens) into one track.
+
Widest coverage, but heterogeneous: compare within a dataset, not across.
+
Best for: does it hold across the single-cell literature?
+
+
+
+
GTEx gives you the organ, the single-cell tracks refine it to the cell type, and the merged track checks whether it holds across many studies.
+
+
+
+
+
Expression: what tissue is it expressed in?
+
+ KLK3 on hg38 with GTEx, Tabula Sapiens, and single-cell tracks: the signal spikes in prostate luminal epithelium and is near-silent elsewhere.
+
KLK3 encodes PSA (prostate-specific antigen), the protein behind the prostate-cancer blood test. Expression is restricted to prostate luminal epithelium, so the contrast is clear.
+
Our thread genes (BRAF, TERT, BRCA2) are broadly expressed, so we pick a textbook tissue-specific gene to make the contrast obvious.
+
Try it
+ Look up KLK3; read its GTEx bars (prostate towers over the rest), then Tabula Sapiens for the cell type. Load the session.
+
+
+
+
+
Regulation & epigenetics
+
enhancers · histone marks · open chromatin · methylation
+
+
+
+
Picking up from “Non-coding SNVs”
+
+
+
+
That set pointed us at GeneHancer, Hi-C/Micro-C, JASPAR and conservation.
+
They all live in the Regulation group, largely from ENCODE.
+
+
Heads-up
+ Two “Regulation” super-tracks exist: ENCODE3 & ENCODE4. Use ENCODE4.
+
Try it, ▶ a real 3D loop at MYC
+ HFFc6 Micro-C: a stripe off the MYC promoter reaches a loop dot in the 8q24 enhancer desert (chr8:128.31–128.33 Mb).
+
+
+ The Regulation group: ENCODE cCREs, DNA Methylation, GeneHancer, Hi-C and Micro-C, JASPAR, VISTA Enhancers and more, all in one place.
+
+
+
+
+
+
Enhancers & promoters: cCREs
+
+
+
+
ENCODE4 cCREs: candidate cis-regulatory elements, on by default.
+ The TERT promoter: the red ENCODE4 cCRE = promoter-like; orange = enhancer-like. Layered H3K27Ac and conservation sit alongside.
+
+
+
+
+
+
TERT: the data say “active promoter” (2/2)
+
+ At the TERT TSS: red promoter cCRE + H3K27Ac + a DNase peak + GeneHancer: every track points to an active promoter.
+
+
cCRE coloured red = promoter-like element
+
H3K27Ac peak = active regulatory region
+
DNase hypersensitivity = open chromatin
+
GeneHancer marks the TERT TSS / regulatory element
+
+
Conclusion
+ Independent assays converge → a real, active promoter, exactly where a non-coding driver mutation bites.
+
+
+
+
+
A variant-interpretation toolkit
+
AI predictors and population frequencies
+
+
+
+
AlphaMissense: AI missense pathogenicity
+
+
+
+
Google DeepMind's deep-learning score for every possible missense substitution (Cheng et al., Science 2023).
+
One value per change, 0 to 1: likely benign → ambiguous → likely pathogenic, drawn at base resolution.
+
The AI counterpart to REVEL: another computational line of evidence (ACMG PP3) for a coding variant.
+
+
Try it
+ Open the AlphaMissense track at BRAF V600 or BRCA2; read the score for our variants.
+
+
+
Why it fits
+ We leaned on REVEL for PP3 earlier. AlphaMissense is the same idea, trained by an AI model, and it covers the whole protein so you can scan a gene for predicted hotspots.
+
+
+
+
+
+
+
SpliceAI: predicting splice disruption
+
+
+
+
Illumina's deep-learning predictor of whether a variant creates or breaks a splice site (Jaganathan et al., Cell 2019).
+
Delta scores (0 to 1) for acceptor / donor gain & loss, with the predicted base position.
+
Lives under the Splicing Impact super-track; SNVs and indels, raw and masked.
+
+
Try it
+ Open Splicing Impact and inspect a splice-region variant.
+
+
+
Closes a loop
+ The ENIGMA demo (Demo 3) cited SpliceAI → PP3 as the evidence that moved a BRCA2 call. This is that track: see the score that drove the reclassification.
+
+
+
+
+
+
+
Variant frequencies: how common, everywhere
+
+
+
+
SNV Frequencies: allele frequencies gathered from population-scale projects worldwide, ~1.7 million genomes / exomes / arrays.
+
One place to compare how common a variant is across populations, ancestries and cohorts, including national projects not in gnomAD.
+
Combined tracks aggregate the data, plus one subtrack per project (TOPMed, gnomAD, 1000 Genomes, and more).
+
note Collected as-is, not re-harmonized: pipelines and assays differ between projects.
+ 
+ 
+ 
+ 
+ 
+ 
+ 
+ 
+ 
+ 
+ 
+ 
+ 
+