To make a STAR reference genome used for alignment you need a genome fasta (the resources fasta provided by the Broad Institute in my case) and a GTF fasta. I tried to use the Ensembl human genome fasta fro hg38 but that does not work. I have looked in the GATK resources package but there is no .gtf file there. I really don't know what to do in this case? Is there anyone who can provide some guidance as to what .gtf file I should use?
$\begingroup$
$\endgroup$
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
3
If you are looking for a genome and its corresponding annotations, you can use the following, it is from nf-core (link to the original doc):
#!/bin/bash
VERSION=108
wget -L ftp://ftp.ensembl.org/pub/release-$VERSION/fasta/homo_sapiens/dna/Homo_sapiens.GRCh38.dna_sm.primary_assembly.fa.gz
wget -L ftp://ftp.ensembl.org/pub/release-$VERSION/gtf/homo_sapiens/Homo_sapiens.GRCh38.$VERSION.gtf.gz
BTW the most up-to-date version as of today is 109.
-
$\begingroup$ When I use the Ensembl it doesn't work saying
Fatal INPUT FILE error, no valid exon lines in the GTF file:and then the path to the Ensembl gtf. $\endgroup$– IndiraCommented Aug 23, 2023 at 11:10 -
$\begingroup$ That makes no sense, Ensembl's GTF is the most comprehensive GTF there is. $\endgroup$– Ram RSCommented Aug 24, 2023 at 16:06
-
$\begingroup$
$\endgroup$
0
I've had success in the past generating STAR indexes using the Gencode GTF files available from here:
https://www.gencodegenes.org/human/
- Comprehensive gene annotation (ALL regions)
- Genome sequence (ALL)
If you make sure that you're getting both the genome and the GTF files from the same source, it's more likely that they will work well together.
Note that the ALL annotation includes assembly patches and alternate haplotypes. The alternate haplotypes mean that population variation is captured, which usually increases the alignment rate of reads.
Unfortunately, most bioinformatics workflows don't properly handle alternate assemblies. As one example, reads aligning to those areas may end up classified as mapping to multiple genomic locations, even though they actually aren't. So it's a tradeoff of a more accurate alignment vs more hassle dealing with those alignments downstream.
