Use of in silico algorithms for variant interpretation - PowerPoint Presentation

Slide1

Use of

in silico

algorithms for variant interpretation

ACMG/AMP Variant

Interpretation guidelines (Richards et al 2015)

Population

Data

In

silico

Data

Segregation

Data

Database

Prevalence in

affecteds statistically increased over controls PS4

MAF frequency is too high for disorder BSI OR observation in controls inconsistent with disease penetrance BS2

Truncating variant in a gene where LOF is a known mechanism of diseasePVS1

De novo (paternity & maternity confirmed) PS2

Well-established functional studies show a deleterious effect PS3

Novel missense change at an amino acid residue where a different pathogenic missense change has been seen before PM5

Multiple lines of computational evidence support a deleterious effect on the gene /gene product PP3

De novo (without paternity & maternity confirmed) PM6

Non-segregation with disease BS4

Patient’s phenotype or FH highly specific for gene PP4

For recessive disorders, detected in trans with a pathogenic variant PM3

Found in case with an alternate cause BP5

Missense in gene where only truncating cause disease BP1

Multiple lines of computational evidence suggest no impact on gene /gene product BP4

Well-established functional studies show no deleterious effect BS3

Located in a mutational hot spot and/or known functional domain PM1

In-frame indels in a repetitive region without a known function BP3

Same amino acid change as an established pathogenic variant PS1

In-frame indels in a non-repeat region or stop-loss variants PM4

Observed in trans with a dominant variant BP2

Functional Data

Co-segregation with disease in multiple affected family members PP1

De novo Data

Allelic Data

Absent in 1000G and

ESP PM2

Strong

Observed in

cis

with a pathogenic variant

BP2

Reputable

source

=

benign

BP6

Strong

Very Strong

Moderate

Supporting

Supporting

Reputable

source

=

pathogenic PP5

Missense in gene with low rate of benign missense variants and path. missenses common PP2

Other Data

Benign

Pathogenic

Increased segregation data

Resources: Table of

in silico toolsAlso see Supplemental data 1 in

http://biorxiv.org/content/early/2017/06/06/146100

and https://omictools.com/functional-predictions-category

Evidence code

Type

In silico tool

BS1

Allele Frequency

https://

cardiodb.org

/

allelefrequencyapp/BS1/PP1/BS4

Allele frequency (1000 genome data + family data integration)

https://sorva.genome.ucla.edu

BS1/PS4

Allele Frequency (specific to cardiovascular disease esp. DCM and ARCV)

https://cardiodb.org/

PM1

Mutation

hotspothttp://

www.cbioportal.orgPM2

Functional domain

1.http://www.uniprot.org

2.http://www.cbioportal.org 3.http://smart.embl-heidelberg.de

/smart/set_mode.cgi?NORMAL=1# 4. www.hrpd.org

PVS1

LOF prediction-variant level

CADD, PROVEAN, Aloft (http://biorxiv.org/content/early/2017/02/07/106468)

PVS1/PP2

LOF prediction-Gene Level

GDI

, LOFTool,pLI, missense Z, local missense Z(http://biorxiv.org

/content/early/2017/06/12/148353), RVIS

PP3/BP4

Missense

see presentation

PP3/BP4

Splicing

see presentation

PP3/BP4

non-coding

GWAVA,Genocanyon

Use of algorithms for interpretations of

1. Missense variants Robert

Huether, Tempus

2. Splicing variants3. Variants impacting 3D structure

Application of

in silico algorithms.

“ If all

of the in silico programs tested agree on the prediction, then this evidence can be counted as supporting. If

in silico predictions disagree, however, then this evidence should not be used in classifying a variant.”

Richards et al, 2015, Genetics in Medicine

ACMG/AMP guideline for use of in silico algorithms

e.g. Polyphen HVAR and HDIV: “HDiv

” is preferred for evaluating rare alleles, dense mapping of regions identified by GWAS, and analysis of natural selection.

“HVar” is better suited for diagnostics of Mendelian diseases, which requires distinguishing mutations with drastic effects from all the remaining

human variation, including abundant mildly deleterious alleles.

Workflow

Annotate variants with 18

in silico

algorithms using dbNSFP 3.2

Liu X, Jian X, and Boerwinkle E. 2013 Human Mutation

Test the applicability of the ACMG/AMP rule for in

silico predictions for missense variants.

14819

missense benign and pathogenic variants from

ClinVar

(≥1 star = rationale for assertion provided by at least one submitter)

Concordance among

in silico

algorithms is low for benign variants

Predictions of algorithms (

pathogenic

/

benign

) for

14819 ClinVar variants using publicly available thresholds

Caveat 1: Concordance among

in silico algorithms is low for benign variantsCaveat 2: False concordances can lead to erroneous inferences.

Example of false concordance:

NM_000059.3(BRCA2):c.4585G>A (p.Gly1529Arg)

Majority of algorithms

assertion are damaging.

Caveat 2: False

concordance increases

with fewer algorithms.

Concordance a across 14819 variants in

ClinVar using publicly available thresholds.

Caveat 3: Some algorithms are sensitive to alignment provided.

https://www.ncbi.nlm.nih.gov/pubmed/21480434

Muscle alignment

SIFT default

Alignment from

Sean Tavtigian

Example with PTEN variants and SIFT

Caveat 4: Combining

conservation metric scores with metapredictors are more likely to give you discordant results.

Summary of some of the caveats

Concordance among algorithms tend to decrease with increasing number of algorithms.

With a smaller set of algorithms there is an increased probability that a variant prediction is opposite to that provided by other lines of evidence.

Sensitivity to alignment

Combining conservation metrics with functional predictors is more likely to give discordant assertions.

Variant may be present in the training set of the algorithm used.Uncalibrated thresholds for pathogenicity cutoff may result in erroneous assertions.

Which predictors perform consistently well across diverse datasets?

Green are

metapredictors

: Crude definition: combines the other predictors

Possible solutions/suggestions

Use gene specific algorithms , if available.Calibrate algorithms to obtain threshold based on known variants in a gene.Separate the evidence into : A) Is it conserved ?B) Is it deleterious by prediction algorithms?

Perhaps best to use these prediction algorithms after exhausting all other sources of evidence.

Given the caveats, it may not be best to elevate PP3/BP4 to a strong or moderate category.Most of the data and summary of the algorithms presented today are here:

http://biorxiv.org/content/early/2017/06/06/146100

Notes on Conservation

PP3/BP4: “The

variant amino acid change being present in multiple nonhuman mammalian species in an otherwise well-conserved region

, suggesting the amino acid change would not compromise function

, can be considered strong evidence for a benign interpretation. One must, however, be cautious about assuming a benign impact in a nonconserved

region if the gene has recently evolved in humans (e.g., genes involved in immune function). “

Good to check the alignment of the region manually. Many ways to do this . One quick way is to use homologene.

1. GERP ,Phylop, (evolutionary constraint) and phastcons (conservation) metrics will give a quantitative output. Note that these are sensitive to alignment.These could be generated from

dbNSFP, VEP or UCSC table browser .For details see associated help pages. e.g. help page in UCSC table browser: http://genome.ucsc.edu

/cgi-bin/hgTrackUi?db=rn6&g=cons20way

2. Manual checking of alignment

Example: DSP gene limited to human disease genes

Alignment

Variant: p.V30M