TRITIUM CLOUD 氚

By Ao Zhang

0%

Posted on Edited on

Introduction to clinical variant interpretation

Scene setting

  • Accurate diagnosis
  • Familial Screening
  • Inform prognosis
  • Personalized treatment

Overview of Rare Variant

1 million variants per person

~1% are protein-coding

Finding one variant that causes a disease is like finding a needle in a haystack.

Overall aim

  • Pathogenic
  • Likely pathogenic
  • Unceartain significance
  • Likely benign
  • Benign

Clinically actionable

For each variant:

  • Have we seen it before?
  • What do we know about it?
  • What can we predict about it?
  • What new data we added?
    • From this family
    • Experiments

ACMG/AMP guidelines

ACMG/AMP guidelines

A Bayesian framework

  • Start with a prior probability
  • Modify it with set of odds, or likelihood
  • Get a posterior probability

Pathogeneic: posterior probability > 0.99
Likely pathogenic: posterior probability > 0.9
Uncertain significance: posterior probability 0.1 - 0.9

Choosing a prior:

  • Variant in small panel testing / large population screening
  • In the range for VUS
  • Use 0.1 (Empirical for BRCA1/2)

Advantages:

  • Incorporate Benign and Pathogenic in one step
  • Can compute odds of pathogenicity for each rule strength to test against functional assays or other metrics
  • More flexible

Evidence

Variants in patients

  • Similar phenotypes

  • How many variants?

  • Patient denominator is often unknown

  • Very unbalanced sample sizes

Use ClinVar to find variants that have been seen before.

Some cautions on using ClinVar data:

  • Bias in certain genes and variants types
  • (Likely) Pathogenic variants are more likely to be submitted
  • Unclear about variant classification
  • Historical 'literature` only curations
  • Data not versioned - classification can change

Functional Data:

  • Matchmaker exchange
  • LitVar
  • PubMed

Segregation Data:

  • Segregation with one or more families

  • Caution:

    • reduced penetrance

De novo occurrence

In controls:

  • Healthy, or population
  • At what frequency?

Allel frequency is one of the strongest predictors of benignity, but is not an indication of pathogenicity.

Population datasets: gnomAD, UK Biobank
Not disease free, expect disease frequency at the same rate as the general population

Allele Frequency (AF)

\[\text{AF} = \frac{\text{number of alleles with variant}}{\text{total number of alleles}}\]

Dominant Disease: < 0.1%

Recessive Disease: < 1%

The proportion of individuals with a variant that develop the disease is the penetrance.

Penetrance:

  • Measure on variant or gene level
  • Often difficult to measure
    • Threshold of disease
    • family based measures are overestimates
    • biobanks likely underestimates
    • variants are often very rare / unique to individual families

Disease-specific statistical approach:

A penetrant dominant mendelian disease

\[\text{Max AF} = \frac{\text{Disease Prevalence} \times \text{heterogeneity}}{\text{Penetrance}}\]

Heterogeneity: maximum proportion of disease attributeable to single variant

Popmax filtering AF (FAF):

Highest allele frequency in any population

if Popmax FAF > Max AF = Too common

Limitations:

  • Low converage: absence does not mean rare
  • Ethnicity: founder variants

Other evidence

  • Highly specific phenotypes that matches gene

  • Observed in trans with known pathogenic allele

  • Similar to previously reported pathogenic variant

Future direction

Moving towards a points based system

Confirming and refuting disease-gene links

GenCC can be used to know the gene and variant that relates to a disease.

PanelApp: Red, Amber, Green

  • Red: No evidence
  • Amber: Some evidence
  • Green: Strong evidence

Experts review the genes and classified to the above categories.

Can be used to determine which variants are diagnostic candidates.

ClinGen: Literature only curation of variants

Burden testing can be used to determine if a gene is associated with a disease

  • But
    • Need large sample sizes
    • Uniform phenotyping
    • Well matched controls
    • Normally limit to rare variants of a particular class
    • Small genes may not be significant
  • Many false disease-gene claims are made

Good Example: MYH7 in dilated cardiomyopathy
significat in ClinVar, Genebass, GenCC

Bad Example: CTNNA3 in ARVC

OMIM disease association
Limited evidence in GenCC and Genebass
Three classification in ClinVar   in gnomAD, many VUSs, limited evidence

Conclusion:

  • Many disease-gene links are not well supported
  • There are many resources to help determine the strength of a disease-gene link

Measures of variant pathogenicity and genetic constraint

Variants with less frequently are more likely damaging.

Modelling mutation rates

To detech genes, regions of genes, or variant types that are under constrained or depleted of variation.

  • Detect shifts in allele frequency spectrum (deleterious variants are rare)

  • Measure a depletion of variants compared to neutral expectation

  • Conservation:

    • Comparison across species (position, region, gene throughout evolution)
    • Requires small number of genomes from different species

  • Constraint:

    • Within human measure
    • Detect Human regions
    • large number of human genomes

???????????????????????????????????????????????????????

Mutability adjusted proportion of singletons (MAPS)

Used to detect the shift in allele frequency spectrum.

More loss-of-function variants should have higher proportions of singletons.

some genetic context is more mutable than others.

Example: C>T at methylated CpG sites

\[\text{MAPS} = \frac{\text{Observed singletons} - \text{Expected singletons}}{\text{Observed variants}}\]

Useful for well-known variant classes

Measuring LoF compared to expected

  • A mutational model for predicting number of variants

  • Gene level constraint metrics

  • LOEUF: rank all genes to tolerance to LoF variants, low score indicates haploinsufficiency.

  • Regional missense constraint

    • MEF2C has a N-terminal DNA binding domain
    • Missense variants in this region cause Developmental Disorders

  • Amino acid level constraint

  • Non-coding constraint

It is useful for dominant early onset phenotypes

Example: BRCA1

But:

  • Ignore some variant contexts
  • Resolution is limited
  • Can't assess individual variants

Individual variant pathogenicity

Majority predicting the deleteroiousness of missense variants

  • SIFT / Mutation assessor / GERP: sequence and evolutionary conservation

  • PolyPhen2: Protein sequence and structure

  • MutPred/Mutation taster: Machine learning

  • REVEL: Ensemble of predictors

Others:

  • SpliceAI: Predicts the effect of variants on splicing
  • CADD: Genome-wide tools

All these tools are supporting level in ACMG/AMP guidelines, meaning they have limited evidence

Generally, REVEL for missense, SpliceAI for splicing, CADD for Genome-wide

Conclusion

Computational tools are useful to determine genes/regions/variant types that likely to be deleterious

Predicting the deleteriousness of underlying data including constraint and cross-species conservation

Used with ACMG/AMP guidelines

Glossary:

  • Dilated Cardiomyopathy: a condition in which the heart becomes weakened and enlarged and cannot pump blood efficiently

  • VUS: Variant of uncertain significance

  • ARVC: Arrhythmogenic right ventricular cardiomyopathy, a genetic disorder that affects the heart muscle

  • haploinsufficiency: a condition in which a single copy of a gene is not sufficient to maintain normal function