MVA Rare Disease Lab

Reproducible genomic analysis for a real child with mosaic variegated aneuploidy. Our Track 1 proposal at the MVA Hackathon 2026 by Sage Bionetworks prioritised a compound heterozygous pair in BUB1B, and it earned the highest score in the track.

Research hypothesis submitted to a hackathon, not a clinical diagnostic report. Variant interpretation and phase confirmation require appropriate clinical validation before any medical use.
See the hypothesis → Method on GitHub ↗
BUB1B / chr15:40161069-40221123 GRCh38
compound heterozygous · autosomal recessive model BubR1 · 1050 aa
5,012,204variants in the patient callset
1 paircompound heterozygous alleles in BUB1B
#1highest score in Track 1
0patient genomic data in the public repository
THE QUESTION

Rare Disease, Real Kid

Behind the data is a real child with mosaic variegated aneuploidy (MVA), a rare disorder of chromosome segregation that combines a predisposition to childhood cancer with severe growth restriction.

Given the patient's genome, which variants explain the disease?
THE HYPOTHESIS

Two BUB1B alleles, one per copy

BUB1B encodes BubR1, a core component of the spindle assembly checkpoint. One allele breaks the protein; the other alters it inside its functional domain: a model consistent with the existing BUB1B and MVA literature.

Allele 1 · truncatingc.2210T>G

p.Leu737Ter

Introduces a stop codon at residue 737, just before the kinase domain, so the whole catalytic domain is lost. Already described as pathogenic and associated in ClinVar with MVA syndrome type 1.

Allele 2 · missensec.3006T>G

p.Asn1002Lys

Ultra-rare: observed once among ~1.46 million alleles in gnomAD v4.1, with no homozygotes. Asparagine to lysine at residue 1002, inside the BubR1 kinase domain.

SIFT 0.000PolyPhen 0.997CADD 26.4
BubR1 protein map: 1050 amino acids, kinase domain at residues 750–1045, stop codon at 737 on allele 1, Asn1002Lys on allele 2. 1 1050 aa ALLELE 1 · p.Leu737Ter STOP at 737 catalytic domain lost ALLELE 2 · p.Asn1002Lys kinase domain 750–1045 Asn→Lys 1002
BubR1 is 1050 amino acids long; its kinase catalytic domain spans roughly residues 750 to 1045.
WHAT WE DID NOT PROVE

The limits, in writing

The VCF carried no phase information and there was no parental segregation data. We could not independently show that the two alleles are in trans, one inherited from each parent.

We left it in the report as an open limitation, together with the validation path that would resolve it. A genomic result that does not state what it did not prove is not a result; it is a claim.

INSIDE THE CELL

From allele to phenotype

  1. 01Two altered alleles

    p.Leu737Ter and p.Asn1002Lys leave BubR1 with reduced function.

  2. 02Weak checkpoint

    The spindle checkpoint no longer holds anaphase until every chromosome is attached.

  3. 03Lagging chromosome

    Division goes ahead anyway and a chromosome is left behind.

  4. 04Unequal daughter cells

    Daughters inherit n+1 and n−1. Repeated through development, this yields a mosaic body.

That is mosaic variegated aneuploidy, and it accounts for the cancer risk and the short stature.

THE FUNNEL

From 5 million variants to one pair

  1. 5,012,204 variants in the genome bcftools view -H "${VCF}" | wc -l
  2. 4,740,790 pass the caller filter bcftools view -f PASS
  3. 14 in the BUB1B locus (13 PASS) bcftools view -r 15:40161069-40221123
  4. 7 PASS heterozygous bcftools view -i 'GT="het"'
  5. 2 exonic on NM_001211.6 (MANE) VEP --offline, MANE transcript
  6. 1 pair the hypothesis presented ClinVar · gnomAD · SIFT · PolyPhen · CADD · domain

Driven by the disease mechanism rather than brute force: prioritise the gene, extract and QC the locus, map to the MANE transcript, and require every candidate to survive converging evidence.

REPRODUCIBLE BY DESIGN

One Makefile, one container

Every target runs inside the container. Whoever clones the repository runs exactly what we ran.

make build        # Ubuntu 24.04 ARM64 + bcftools, samtools, bedtools
make check        # every tool answers before touching data
make qc           # quality control of the full VCF
make bub1b        # targeted extraction of the locus
make bub1b-phase  # phase review of the candidates

# Functional annotation, offline against a local cache
vep --cache --offline --dir_cache /vep_data \
    --species homo_sapiens --assembly GRCh38 \
    --symbol --canonical --mane --sift b --polyphen b \
    --af --af_gnomade --af_gnomadg --max_af --clin_sig

# Last step: audit before the repository goes public
bash scripts/prepublish_audit.sh
PRIVACY BY DESIGN

Anyone can reproduce the method

Local

Runs on our machine

The whole analysis ran locally in a reproducible container. --offline guarantees no patient coordinate leaves the machine to be resolved by an external service.

Public

Method, not data

The public repository holds the Dockerfile, scripts, pipeline and documentation, and no patient genomic data: no raw VCFs, genotype-level tables or downloaded references.

Audited

A script, not a checklist

An audit script runs before the repository is made public. Checking by hand is not a control; a script that fails is.

github.com/ivillafuerte/mva-hackathon-2026 ↗ Read the full article →

Build something great.

Start deploying Contactar a ventas