Variant Annotation Integrator (2024)

Introduction

The Variant Annotation Integrator (VAI) is a research tool for associatingannotations from the UCSC database with your uploaded set of variant calls.It uses gene annotations to predict functional effects of variants on transcripts.For example, a variant might be located in the coding sequenceof one transcript, but in the intron of an alternatively spliced transcriptof the same gene; the VAI will return the predicted functional effectfor each transcript. The VAI can optionally add several othertypes of relevant information: the dbSNP identifier if the variantis found indbSNP,protein damage scores for missense variants from theDatabase of Non-synonymous Functional Predictions (dbNSFP),and conservation scores computed from multi-species alignments.The VAI can optionally filter results to retain only specific functionaleffect categories, variant properties and multi-species conservation status.

NOTE:
The VAI is only a research tool, meant to be used by those who have beenproperly trained in the interpretation of genetic data,and should never be used to make any kind of medical decision.We urge users seeking information about a personal medical or geneticcondition to consult with a qualified physician for diagnosis and foranswers to personal questions.

Submitting your variant calls

In order to use the VAI, you must provide variant calls in either thePersonal Genome SNP (pgSnp) orVCF format.pgSnp-formatted variants may be uploaded as aCustom Track.Compressed and indexed VCF files must be on a web server (HTTP, HTTPS or FTP)and configured as Custom Tracks, or if you happen to have aTrack Hub,as hub tracks.

Protein-coding gene transcript effect predictions

Any gene prediction track in the UCSC Genome Browser database or in a track hubcan be selected as the VAI's source of transcript annotations for predictionof functional effects.Sequence Ontology (SO) terms are used to describe the effectof each variant on genes in terms of transcript structure as follows:

Optional annotations

Database of Non-synonymous Functional Predictions (dbNSFP)

dbNSFP provides scores and predictions from several tools that use variousmachine learning techniques to estimate the likelihood that a single-nucleotidemissense variant would damage a protein's structure and function:

• SIFT (Sorting Intolerant From Tolerant) uses sequence hom*ology and the physical properties of amino acids to predict whether an amino acid substitution affects protein function. Scores less than 0.05 are classified as Damaging ("D" in output); higher scores are classified as Tolerated ("T"). (Ng and Henikoff, 2003)
• PolyPhen-2 (Polymorphism Phenotyping v2) applies a naive Bayes classifier using several sequence-based and structure-based predictive features including refined multi-species alignments. PolyPhen-2 was trained on two datasets, and dbNSFP provides scores for both. The HumDiv training set is intended for evaluating rare alleles potentially involved in complex phenotypes, for example in genome-wide association studies (GWAS). Predictions are derived from scores, with these ranges for HumDiv: "probably damaging" ("D") for scores in [0.957, 1]; "possibly damaging" ("P") for scores in [0.453, 0.956]; "benign" ("B") for scores in [0, 0.452]. HumVar is intended for studies of Mendelian diseases, for which mutations with drastic effects must be sorted out from abundant mildly deleterious variants. Predictions are derived from scores, with these ranges for HumDiv: "probably damaging" ("D") for scores in [0.909, 1]; "possibly damaging" ("P") for scores in [0.447, 0.908]; "benign" ("B") for scores in [0, 0.446]. (Adzhubei et al., 2010)
• MutationTaster applies a naive Bayes classifier trained on a large dataset (>390,000 known disease mutations from HGMD Professional and >6,800,000 presumably harmless SNP and Indel polymorphisms from the 1000 Genomes Project). Variants that cause a premature stop codon resulting in nonsense-mediated decay (NMD), as well as variants marked as probable-pathogenic or pathogenic in ClinVar, are automatically presumed to be disease-causing ("A"). Variants with all three genotypes present in HapMap or with at least 4 heterozygous genotypes in 1000 Genomes are automatically presumed to be harmless polymorphisms ("P"). Variants not automatically determined to be disease-causing or polymorphic are predicted to be "disease-causing" ("D") or polymorphisms ("N") by the classifier. Probability scores close to 1 indicate high "security" of the prediction; probabilities close to 0 for an automatic prediction ("A" or "P") can indicate that the classifier predicted a different outcome. (Schwarz et al., 2010)
• MutationAssessor uses sequence hom*ologs grouped into families and sub-families by combinatorial entropy formalism to compute a Functional Impact Score (FIS). It is intended for use in cancer studies, in which both gain of function and loss of function are important; the authors also identify a third category, "switch of function." A prediction of "high" or "medium" indicates that the variant probably has some functional impact, while "low" or "neutral" indicate that the variant is probably function-neutral. (Reva et al., 2011)
• LRT (Likelihood Ratio Test) uses comparative genomics to identify variants that disrupt highly conserved amino acids. Variants are predicted to be deleterious ("D"), neutral ("N") or unknown ("U"). (Chun and Fay, 2009)
• VEST (Variant Effect Scoring Tool) (available only for hg38/GRCh38) uses a classifier that was trained with ~45,000 disease mutations from HGMD and ~45,000 high frequency missense variants (putatively neutral) from the Exome Sequencing Project. (Carter et al., 2013)

In addition, dbNSFP providesInterPro protein domains where available(Hunter et al., 2012) and two measuresof conservation computed byGERP++(Davydov et al., 2010).

Transcript status

Some of the gene prediction tracks have additional annotations to indicate the amount or quality of supporting evidence for each transcript. When the track selected in the "Select Genes" section has such annotations, these can be enabled under "Transcript Status". The options depend on which gene prediction track is selected.

• GENCODE tags: when GENCODE Genes are selected in the "Select Genes" section, any GENCODE tags associated with a transcript can be added to output.
• RefSeq status: when RefSeq Genes are selected in the "Select Genes" section, the transcript's status can be included in output.
• Canonical UCSC transcripts: when UCSC Genes (labeled GENCODE V22 in hg38/GRCh38) are selected in the "Select Genes" section, the flag "CANONICAL=YES" is added when the transcript has been chosen as "canonical" (see the "Related Data" section of the UCSC Genes track description).

Known variation

If the selected genome assembly has a SNPs track (derived fromdbSNP),when a variant has the same start and end coordinates as a variant indbSNP,the VAI includes the reference SNP (rs#) identifier in the output.Currently, the VAI does not compare alleles due to the frequency of strandanomalies in dbSNP.

Conservation

If the selected genome assembly has a Conservation track with phyloP scoresand/or phastCons scores and conserved elements,those can be included in the output.Both phastCons and phyloP are part of thePHAST package;see the Conservation track description in the Genome Browserfor more details.

Filters

The volume of unrestricted output can be quite large,making it difficult to identify variants of particular interest.Several filters can be applied to keep only those variantsthat have specific properties.

Functional role

By default, all variants are included in the output regardless ofpredicted functional effect. If you would like to keep only variantsthat have a particular type of effect, you can uncheck the checkboxesof other effect types.The detailed functional effect predictions are categorized as follows:

• intergenic: intergenic_variant
• upstream/downstream of gene: upstream_gene_variant, downstream_gene_variant
• 5' or 3' UTR: 5_prime_UTR_variant, 3_prime_UTR_variant
• CDS - synonymous coding change: synonymous_variant
• CDS - non-synonymous (missense, stop gain/loss, frameshift etc): missense_variant, inframe_insertion, inframe_deletion, frameshift_variant, initiator_codon_variant, incomplete_terminal_codon_variant, stop_lost, stop_retained_variant, stop_gained, NMD_transcript_variant
• intron: intron_variant
• splice site or splice region: splice_donor_variant, splice_acceptor_variant, splice_region_variant
• exon of non-coding gene: non_coding_exon_variant

Known variation

(applicable only to assemblies that have "Common SNPs" and"Mult SNPs" tracks)By default, all variants appear in output regardless of overlap with knowndbSNP variants that map to multiple locations (a possible red flag),or that have a global minor allele frequency (MAF) of 1% or higher.Those categories of known variants can be used to exclude overlapping variantsfrom output by unchecking the corresponding checkbox.

Conservation

(applicable only to assemblies that have "Conservation" tracks)If desired, output can be restricted to only those variants that overlapconserved elements computed by phastCons.

Output format

Currently, the VAI produces output comparable to Ensembl'sVariant Effect Predictor (VEP), in either tab-separatedtext format or HTML.Columns are describedhere.When text output is selected, entering an output file name causes output tobe saved in a local file instead of appearing in the browser, optionallycompressed by gzip (compression reduces file size and network traffic,which results in faster downloads).When HTML is selected, output always appears in the browser window and theoutput file name is ignored.

Acknowledgments

Anyone familiar with Ensembl'sVariant Effect Predictor (VEP) will doubtless noticesimilarities in options and interface. In collaboration with our colleaguesat Ensembl, we have made an effort to limit the differences between the toolsby using Sequence Ontology terms to describe variants' functional effects andby creating a "VEP" output format.Any bugs in the VAI, however, are in the VAI only.