Genome Annotation: Genes and Variant Effects

A newly sequenced genome is just a very long string of the letters A, C, G, and T, with nothing marking what any stretch actually means. Genome annotation adds that missing layer of meaning, and it breaks into two linked tasks that this lesson walks through.

Structural And Functional

Structural annotation locates features on the sequence, such as a gene (a stretch of DNA carrying the instructions for a product) and its exons (the pieces that stay in the final messenger RNA after the in-between introns are cut out). Functional annotation then labels what each feature does, for example that a gene encodes an oxygen-carrying protein. You always do structural annotation first, because you cannot describe a feature you have not yet found.

Gene Models In GFF3

Annotations are stored in plain-text feature tables, most often in the GFF3 format or the closely related GTF format. Each line describes one feature using nine tab-separated columns: the sequence name, the source, the feature type, the start and end positions, a score, the strand, a phase, and a final column of attributes. Features are nested into a gene model, where a gene contains one or more transcripts (mRNA), and each transcript contains exons and coding sequences (CDS).

##gff-version 3
chr1	Ensembl	gene	1000	9000	.	+	.	ID=gene:ENSG01;Name=HBB
chr1	Ensembl	mRNA	1000	9000	.	+	.	ID=tx:ENST01;Parent=gene:ENSG01
chr1	Ensembl	exon	1000	1200	.	+	.	Parent=tx:ENST01
chr1	Ensembl	exon	7000	9000	.	+	.	Parent=tx:ENST01
chr1	Ensembl	CDS	1050	1200	.	+	0	Parent=tx:ENST01

Variant Effect Prediction

Once genes are annotated, you can ask what a genetic variant does to them. Tools like Ensembl VEP (Variant Effect Predictor) and SnpEff read a list of variants and compare each one against the gene models to predict its consequence. Common consequences include a synonymous_variant (the DNA changes but the protein stays the same), a missense_variant (one amino acid is swapped for another), and a stop_gained variant (a change creates an early stop codon that cuts the protein short).

The command below runs VEP on an input VCF file, the standard text format that lists variant positions and their alleles, and writes the annotated results to an output file. The --cache and --offline flags tell VEP to use a pre-downloaded local database instead of querying the internet, while --species homo_sapiens picks the human gene set to compare against. The --tab flag asks for tab-separated output, and --symbol adds the human-readable gene name to each result. VEP's real output carries more columns than are shown below, but these few are enough to read off each variant's effect.

vep --input_file variants.vcf --output_file variants.vep.txt --species homo_sapiens --cache --offline --tab --symbol

#Uploaded_variation	Location	Allele	Gene	SYMBOL	Consequence	Amino_acids	Codons
var1	7:140753336	T	ENSG00000157764	BRAF	missense_variant	V/E	gTg/gAg
var2	11:5226774	T	ENSG00000244734	HBB	synonymous_variant	L	ctG/ctA
var3	17:7673776	A	ENSG00000141510	TP53	stop_gained	Q/*	Cag/Tag