Variant calling

Variant calling is the process of looking for differences in the genome. This could be looking for differences between individuals (germline calling) or between a tumor and matched normal tissue (somatic calling).

  • Koboldt 2020 is a good overview of variant calling, including different technologies, differences between germline and somatic calling, and a sketch of the workflow. Figure 2 is especially good to study as it has examples of incorrectly called variants and shows how to critically evaluate calls.

  • Data Carpentry’s “Wrangling Genomics” workshop has a section on variant calling that walks through some of the basics. Depending on which machine you’re running it on, you may want to start at the beginning of the workshop which walks you through installing the required software. If this is your first time working on the command line, then the entire workshop would be good to go through.

  • The Broad’s Genome Analysis Toolkit (GATK) is the standard germline variant calling workflow (often called “GATK best-practices workflow”) that can also do somatic calling and germline CNV. That site includes extensive tutorials and blog posts that give a lot of detail and context.

Terminology

SNV:

Single-nucleotide variant, typically use GATK to call these.

indel:

Insertion or deletion. Typically assumed to be small-ish (<10 bp), called by GATK.

SV:

Structural variant. Typically assumed to be large (>1kb). Can include complicated inversions, translocations, and copy-number variants (which are often considered a separate class). SVs are best detected with long-read sequencing and need specialized algorithms to detect them. This is a hard problem, and different algorithms are able to detect different kinds of SVs.

CNV:

Copy number variants. This is a stretch of DNA that is duplicated. Often assayed by aCGH (array comparative genome hybridization).

germline variant calling:

The process of identifying places in a genome that differ from the reference genome. If otherwise unspecified, it usually means SNV and indel calling (rather than SV or CNV).

somatic variant calling:

Calling variants that arose in a tumor compared to normal tissue is somatic varant calling. It is important to note that somatic calling is NOT simply a difference between two callsets, it’s quite a bit more complicated than that.

Interpreting and prioritizing variants

Calling variants is just the first step. You’ll probably find thousands. Which ones are meaningful?

Variant annotation is the process of adding extra information to each variant. For example, the location of each variant relative to genes, known clinical relevance, and how common a variant is in the general population.

Dashti 2018 provides lots of practical advice about tools, databases, and thresholds to use and is worth a careful read.

Example tools for this include:

There are many databases, each with their own uses. There is no single database that aggregates them all, though dbNSFP comes the closest. Here is an incomplete list:

  • dbNSFP compiles prediction scores for non-synonymous single-nucleotide variants for many databases (some of which are also listed below). For one-off queries, there is a web application. This is probably the best one-stop-shop for variant annotation. But when digging deeper on a small number of variants, you may find it useful to submit those variants to the tools

    > Its current version is based on the Gencode release 29 / Ensembl version 94

    and includes a total of 84,013,490 nsSNVs and ssSNVs (splicing-site SNVs). It compiles prediction scores from 37 prediction algorithms (SIFT, SIFT4G, Polyphen2-HDIV, Polyphen2-HVAR, LRT, MutationTaster2, MutationAssessor, FATHMM, MetaSVM, MetaLR, CADD, CADD_hg19, VEST4, PROVEAN, FATHMM-MKL coding, FATHMM-XF coding, fitCons x 4, LINSIGHT, DANN, GenoCanyon, Eigen, Eigen-PC, M-CAP, REVEL, MutPred, MVP, MPC, PrimateAI, GEOGEN2, BayesDel_addAF, BayesDel_noAF, ClinPred, LIST-S2, ALoFT), 9 conservation scores (PhyloP x 3, phastCons x 3, GERP++, SiPhy and bStatistic) and other related information including allele frequencies observed in the 1000 Genomes Project phase 3 data, UK10K cohorts data, ExAC consortium data, gnomAD data and the NHLBI Exome Sequencing Project ESP6500 data, various gene IDs from different databases, functional descriptions of genes, gene expression and gene interaction information, etc.

  • dbSNP is hosted by NCBI and is a collection of known SNPs corresponding with an rsID, a unique identifier. If your variant calling found variant in dbSNP, annotation tools will generally annotate it with the rsID.

  • dbVar is similar to dbSNP but for structural variation

  • ClinVar stores clinical information about variants

  • ClinGen is a curated resource of clinically-relevant variants in genes (read more about the differences between ClinVar and ClinGen here)

  • OMIM lists known genetic disorders and associated genes.

  • SIFT (“Sorting Intolerant from Tolerant”). More sophisticated than just “amino acid change”, SIFT predicts whether an amino acid change would affect protein function based on sequence homology and physical properties of amino acids.

  • PolyPhen (PolyPhen2-HDIV and PolyPhen2-HVAR are included in dbNSFP). dbNSFP includes SIFT, but only for known cases. Unknown missense mutations can be run through the command-line tool to calculate predictions.

  • COSMIC is the Catalogue of Somatic Mutations in Cancer which is an expert-curated database of somatic mutations. The website also has mutation profiles of 1000+ cell lines and other useful info on somatic mutations in human cancer. You will need to create an account to download data.

  • GERP (Genomic Evolutionary Rate Profiling) assesses if a mutation was likely to be neutral or not (precomputed scores included in dbNSFP)

  • CAAD (Combined Annotation Dependent Depletion) scores the deleteriousness of single-nucleotide variants and small indels. The scoring uses a combination of sources. Precomputed scores are included in dbNSFP.

Visualization

The gnomAD browser is a great way to visualize your variants.

The UCSC Genome Browser has tracks for many of the databases noted above.

Krusche 2019 in Figure 2 show the ways in which small indels can be called in different ways.

Additional notes

Todo

This section could use some better organization

Filtering on minor allele frequency: A non-synonymous SNP that is rare in the general population might be expected to be more important for the study. Annotation tools typically add MAF (minor allele frequency) which is taken from large studies with healthy individuals. But that “healthy individuals” assumption is probably not 100% correct, due to incomplete penetrance or variable clinical phenotype. So while we might initially think of something like 1% as a threshold for “rare in the general population”, we should set that possibly to some different value. One place to start would be to set the threshold to the carrier frequency for a related disease.

Some notes from from MacArthur 2018:

How many samples needed:

> Gene discovery for conditions with low locus heterogeneity and sufficiently

high-penetrance mutations is occasionally possible by sequencing a single family; however, most gene-discovery applications will require substantially larger sample sizes: multiple unrelated families for rare monogenic conditions, and thousands to tens of thousands of patients and controls for complex disorders.

If you want to say that you did not find an expected variant:

> Investigators should begin by examining sequence variation in genes known to

be associated with that phenotype, and assessing sequence coverage of the coding sequences and splice junctions for these genes before exploring the possibility of new candidate genes in the affected individuals.

A tale of caution regarding gene size:

> A study found 4 independent missense mutations in TTN in a cohort of 945

families with a child affected by autism. TTN has the largest coding sequence of any gene, so you’d actually expect 1.96 mutations. Need to consider gene size, mutation rate, number of trios, distribution of exome coverage.

  • Eilbeck 2017 gives a good overview of variant prioritization in the context of Mendelian disease.

  • MacArthur 2014 gives clear guidelines assessing variant pathogenicity.

  • The ACGM standards describes standard specific terminology like “pathogenic” or “likely benign” to describe variants in genes causing Mendelian disorders.

Other resources