Extra: Building gene bed files from a GTF: dynema-prepare-bed

dynema-map --bed is driven by a small bed-like gene file – one row per gene with chr, start, end, gene, strand (tab-separated), from which the TSS is derived FastQTL-style (start on the + strand, end on the - strand). For a handful of genes it is easiest to write by hand:

chr1	150730706	150765778	CTSS	-
chr11	2323216	2339430	TSPAN32	+

Gene identifiers are matched against the expression features file by gene name first, then gene id – so CTSS and ENSG00000163131 both work, and Ensembl .N version suffixes are ignored.

For a genome-wide run, build the file(s) straight from your GTF annotation (GENCODE, Ensembl, or the Cell Ranger reference's genes.gtf; plain or gzipped), matched against the expression data's features file:

./bin/dynema-prepare-bed --gtf genes.gtf.gz --features expr.features.gz \
  --chunk-size 100 --out beds/chunk

With --chunk-size 100 this writes beds/chunk_001.bed, beds/chunk_002.bed, ... of at most 100 genes each, in GTF order (so chunks are genomically contiguous) – the intended genome-wide pattern is then one dynema-map job per chunk file on your HPC scheduler, with --check and --skip-existing for safe submission and resubmission (see batching). Omit --chunk-size (or pass 0) to write everything to a single <out>.bed.

For each GTF gene record the tool picks the identifier that dynema-map will resolve unambiguously at map time:

• the gene name/symbol, when it is unique both within the GTF and in the features file (dynema-map searches feature gene names first);
• the gene id otherwise (symbol absent, or duplicated in the features file or annotated at multiple loci in the GTF);
• genes not present in the features file at all are dropped – they could never be mapped anyway.