Annotation generation bug fixes

[ainefairbrother]

SpliceAI input processing bug fixes

Summary: off-by-one and multi-transcript handling fixes for annotation file creation and the variant simulation tool.

Description of changes

ensembl-variation/scripts/python/spliceai_annotation_file.py fixes:

1. In the gff3 branch, currently gff3 tx and exon starts are written to the annotation file as-is, however, this is not correct, as SpliceAI adds +1 to tx start and exon start when importing and processing the annotations. As such, SpliceAI assumes that the annotation file has 0-based starts and 1-based ends, and to achieve this, starts should get -1 and ends should remain unchanged.
2. Handling of multiple transcripts - as per the SpliceAI author, this should be one gene-tx per row in the annotation file.
3. Other small fixes (robustifying, extra filters, remove buggy functionality):
   • Filter for main chromosomes only (1-22, X, Y),
   • Name cleaning/ generating functions (for gene:tx name).

The annotation file now looks like:

#NAME	CHROM	STRAND	TX_START	TX_END	EXON_START	EXON_END
TOP1:ENST00001107672	20	+	41028814	41033491	41028814,41029430,	41029100,41033491,
APP:ENST00001107674	21	-	25910135	26170830	25910135,25954589,25955626,25975069,25975953,25982343,26000014,26021839,26050999,26053235,26089942,26111978,26170563,	25911962,25954689,25955755,25975228,25976028,25982477,26000182,26022042,26051193,26053348,26090072,26112146,26170830,
APP:ENST00001107675	21	-	25910144	26170830	25910144,25954589,25955626,25975069,25975953,25982343,25997359,26000014,26021839,26050999,26053235,26089942,26111978,26170563,	25911962,25954689,25955755,25975228,25976028,25982477,25997416,26000182,26022042,26051193,26053348,26090072,26112146,26170830,

tools/variant_simulator/simulate_variation fixes:

1. In the onlyGencodePrimary, add all GENCODE-primary protein-coding transcript to the output list, rather than just the final one.

Removal of the scripts/python/filter_annotation_file.py script - this was for the incorrect exon diff-ing approach.

Testing

You can generate and inspect the GENCODE primary protein coding annotation file using:

python3 spliceai_annotation_file.py \
  --gff3 Homo_sapiens.GRCh38.116.gff3.gz \
  --gencode_primary \
  --name_format gene_transcript \
  --release 116 \
  --output_file "gene_annotation_116_gencode_primary_protein_coding.txt"

[dglemos]

Tested against file Homo_sapiens.GRCh38.115.chr.gff3.gz

• spliceai_annotation_file.py generates the expected output taking into account the start coordinate being 0-based
• simulate_variation returns all regions overlapping gencode transcripts

[dglemos]

Can you also update this example with the correct start coordinates (-1)?

[dglemos]

This comment is in line 21.

[dglemos]

You could also add some information about the chrs included in the output file, something like this:

In both modes, only chromosomes 1-22, X, and Y are retained; all other contigs (including MT) are excluded from the generated annotation file.