Rapid and robust estimation of nucleosome-free regions and nucleosome positioning.
Most of the core functionality in fastocc is accessible through the CLI with a small
number of commands.
The bulk of processing is done by the call command, which takes a .bam file containing
paired ATAC-seq reads, processes it into fragments, and then counts those fragments into
nucleosomal (longer) and nucleosome-free (shorter) bins along the genome (default is 10bp).
fastocc can be run in tiles over the whole genome, but the algorithm will generally
be uninformative outside of regions of high ATAC-seq signal (ie peaks), specified with
the --bed argument.
fastocc call --bam my_atac.bam --bed peaks.bed --output fragments.h5wig
.h5wig is a novel format (built on HDF5), which stores a set of regions (e.g. peaks),
and acro each region tracks multiple signals in consecutive bins across the whole region.
This concept of binned data in ranges is used internally in many places in fastocc,
and is a significant reason behind its efficiency.
We can then call nucleosome-free regions using the nfr.
fastocc nfr --fragments fragments.h5wig --output nfrs.bed
If we want to visualise the predicted occupancy in a genome browser (or the counts of
NF for nucleosomal reads), we can export these from the fragments file as a .bigWig:
fastocc export --fragments fragments.h5wig --track occ --output occ.bw
Lastly, we can export everything to an extended .bedGraph format. This isn't recommended
for the whole dataset due to the resulting file size, so a set of regions can be provided:
fastocc dump --fragments fragments.h5wig --bed regions_of_interest.bed --output fragments.bg
If you do need programmitic random access to the fragment-level data, functions are
provided in Python (exported in the package) and R (in extra/). Implementing reading
should be straightforward in any other language with a mature HDF5 library.
More information about the package can be found in the documentation and source code, and the Jupyter notebooks included in the repository.
Some things you might be interested in:
- Merge NFRs across cell lines to generate precises consensus cis-regulatory elements.
- Scan NFRs for enriched motifs.
- Predict TF-binding based on ChIP-seq and NFRs
- Use the internal API to analyse scATAC data directly from
.arrowfiles.
If you do any of this, please let me know! Github DMs are the best way to reach me, see here. If you have any issues or feature requests, I will try to watch issues here but may require a nudge. I am no longer working on this full time, so you may be asked to contribute, please don't be scared to do this!
The development of fastocc took place within the
Ensembl Regulation team at
EMBL-EBI. The project received funding from the European Union’s Horizon 2020
programme under the AQUA-FAANG
and GENE-SWitCH projects.
The tool itself was based on ideas originally implemented in the NucleoATAC package, if
you're interested in the biological reasoning behind fastocc you sohuld read their
paper in Genome Research and check out
NucleoATAC on GitHub.
fastocc is available under the Apache v2.0 license, included in the repository.
