Didn't find make_genome_memmap.py script

[JiayiJennie]

Hi!

This is an interesting work!

When reproducing results, I want to find data/promoter_design/make_genome_memmap.py to generate .mmap file for human reference when generating promoter sequence, but didn't find it. Thanks!

[raghuramdr]

Hi, great work. Seconding the original question. The file data/promoter_design/make_genome_memmap.py is not present. Can you please upload it to reproduce the results? Cheers!

[PavelAvdeyev]

Dear all,

Solution 1

I added make_genome_memmap.py file into directory external.py. make_genome_memmap.py requires selene_utils.py file. The usage of this file is the following:

python make_genome_memmap.py <path_to_fasta>

The script creates file with one-hot encoding of a reference genome provided in fasta file.

For more details, one can visit this repository:

To speed up genome sequence retrieval, the training scripts use a memory-mapped genome file. Run python misc/make_genome_memmap.py before you use the training scripts (47Gb space needed).

Solution 2

Alternative solution is not to use a memory-mapped genome. For doing so, one can replace code of genome dataloader from:

        self.genome = MemmapGenome(
            input_path=config.ref_file,
            memmapfile=config.ref_file_mmap,
            blacklist_regions='hg38'
        )

to

         self.genome = MemmapGenome(
            input_path=config.ref_file,
            memmapfile=None,
            blacklist_regions='hg38'
        )

See lines 140-143 in file: train_promoter_designer.py

Hopefully, it helps.