BED, GFF & GTF

BED, GFF and GTF describe regions and annotations on a genome. BED defines intervals such as peaks or capture regions. GFF and GTF define gene models with exons, introns and coding sequences. You meet all three regularly.

BED format

BED is a tab-separated format for genomic intervals. The minimum file has three columns: chromosome, start and end.

chr1    1000    2000
chr1    3000    3500
chr2    5000    6000

BED uses zero-based, half-open coordinates. The start is 0-based and the end is excluded, so chr1 1000 2000 covers bases 1001 to 2000 in 1-based counting. This is the single most common source of off-by-one errors in bioinformatics.

Caution

VCF and SAM use 1-based coordinates and BED uses 0-based. Converting between them means adding or subtracting 1 from the start. Get it wrong and every region shifts by one base.

BED3, BED6 and BED12

BED supports extra columns beyond the required three.

Variant	Columns	Extra fields
BED3	3	chrom, start, end only
BED6	6	adds name, score, strand
BED12	12	adds thickStart, thickEnd, itemRgb, blockCount, blockSizes, blockStarts

A BED6 file looks like this:

chr1    1000    2000    peak_1    500    +
chr1    3000    3500    peak_2    300    -
chr2    5000    6000    peak_3    800    +

The name labels the region, the score carries a numeric value such as a peak signal, and the strand is + or -. BED12 adds a block structure that represents exons within a transcript through the blockCount, blockSizes and blockStarts fields.

You meet BED files everywhere: MACS2 writes ChIP-seq and ATAC-seq peaks as BED, exome capture kits ship a target BED that variant callers use to restrict the search space, the ENCODE blacklist is a BED of artefact-prone regions to exclude, and any custom set of regions of interest travels as BED.

GFF3 format

GFF3 describes gene structures and other genomic features in nine tab-separated columns.

##gff-version 3
chr1    ensembl    gene       1000    5000    .    +    .    ID=gene1;Name=BRCA1
chr1    ensembl    mRNA       1000    5000    .    +    .    ID=tx1;Parent=gene1
chr1    ensembl    exon       1000    1200    .    +    .    Parent=tx1
chr1    ensembl    CDS        1050    1200    .    +    0    Parent=tx1
chr1    ensembl    exon       3000    3300    .    +    .    Parent=tx1
chr1    ensembl    CDS        3000    3300    .    +    2    Parent=tx1

Column	Name	Description
1	seqid	Chromosome or scaffold name
2	source	Program or database that produced the feature
3	type	Feature type: gene, mRNA, exon, CDS
4	start	1-based start position
5	end	1-based end position, inclusive
6	score	Numeric score or .
7	strand	+ or -
8	phase	For CDS: 0, 1 or 2, the reading frame offset
9	attributes	Semicolon-separated key=value pairs

GFF3 is hierarchical. A gene contains mRNA transcripts, and each transcript contains exons and CDS features. The Parent attribute links a child feature to its parent. GFF3 uses 1-based, fully closed coordinates: both the start and the end base belong to the feature.

GTF format

GTF is a close relative of GFF with a different attribute syntax, and it is the standard input for RNA-seq tools.

chr1    ensembl    gene       1000    5000    .    +    .    gene_id "ENSG00000012048"; gene_name "BRCA1";
chr1    ensembl    transcript 1000    5000    .    +    .    gene_id "ENSG00000012048"; transcript_id "ENST00000357654";
chr1    ensembl    exon       1000    1200    .    +    .    gene_id "ENSG00000012048"; transcript_id "ENST00000357654"; exon_number "1";

The difference sits in column 9. GTF uses key "value"; pairs, GFF3 uses key=value. GTF requires the gene_id and transcript_id attributes. GTF also uses 1-based, fully closed coordinates.

GFF or GTF

Most RNA-seq alignment and quantification tools expect GTF. STAR, HISAT2, featureCounts and Cell Ranger all take GTF annotations, and GENCODE ships GTF by default. GFF3 is more common at NCBI and in non-model organism databases, and Ensembl provides both. When a tool asks for a gene annotation file, check its documentation, but in most cases it wants GTF.

Tip

Convert between GFF3 and GTF with gffread or agat. Verify the output after conversion, because edge cases fail silently.

Where to download annotations

Source	Site	Formats	Notes
GENCODE	gencodegenes.org	GTF, GFF3	Human and mouse. The standard for RNA-seq.
Ensembl	ensembl.org	GTF, GFF3	Many species, Ensembl gene IDs.
UCSC	genome.ucsc.edu	Various	Table Browser for custom downloads, chr prefix.
NCBI	ncbi.nlm.nih.gov	GFF3	RefSeq annotations, broad species coverage.

Caution

Chromosome naming differs between sources. GENCODE and UCSC write chr1, Ensembl writes 1. The annotation must match the chromosome names in the reference genome and the BAM files, or tools count zero reads per gene and say nothing.

bedtools basics

bedtools is the standard toolkit for interval operations, and it reads BED, GFF, GTF, VCF and BAM.

Intersect, merge, closest

# Find peaks that overlap promoter regions
bedtools intersect -a peaks.bed -b promoters.bed > overlapping_peaks.bed

# Report the peak and the overlapping promoter together
bedtools intersect -a peaks.bed -b promoters.bed -wa -wb > peaks_with_promoters.bed

# Keep peaks that do NOT overlap blacklist regions
bedtools intersect -a peaks.bed -b blacklist.bed -v > filtered_peaks.bed

# Merge overlapping peaks (merge needs sorted input)
sort -k1,1 -k2,2n peaks.bed > peaks_sorted.bed
bedtools merge -i peaks_sorted.bed > merged_peaks.bed

# Merge peaks within 500 bp of each other
bedtools merge -i peaks_sorted.bed -d 500 > merged_peaks_500bp.bed

# The nearest gene for each peak
bedtools closest -a peaks.bed -b genes.bed > peaks_nearest_gene.bed

The -a flag takes the query file and -b the reference file. The -v flag inverts the match and reports entries with no overlap.

Other frequent operations

# Remove blacklisted areas from peaks
bedtools subtract -a peaks.bed -b blacklist.bed > clean_peaks.bed

# 1000 bp upstream of each gene
bedtools flank -i genes.bed -g genome.sizes -l 1000 -r 0 > upstream_1kb.bed

Tip

Many bedtools operations need sorted input. Sort BED files with sort -k1,1 -k2,2n file.bed before merge, complement or genomecov.

Summary

• BED defines intervals with 0-based, half-open coordinates, in 3 to 12 columns.
• GFF3 describes hierarchical gene models with 1-based coordinates and is common at NCBI and Ensembl.
• GTF is the annotation format RNA-seq tools expect. Download it from GENCODE for human and mouse work.
• Coordinate systems differ between formats. Always check which one you hold.
• bedtools covers interval work: intersect, merge and closest handle most of it.
• Chromosome names must match between annotation, reference genome and BAM files.