ATAC-seq Data Integration

[ilan-gold]

Cell by bin
Visualize in higlass
Cell by peaks (in BED + snap files)
Annotated peaks (genomic intervals) per cell
Genome-wide (not necessarily tied to a gene)
Our Pipelines:

• We need to create all the new pipelines.

TMC:

• CalTech
• Stanford

Outstanding Issues:

• Creation of new pipelines

Notes:
We have many visualization options:

1. Cell by gene

• A result of the alignment? Perhaps less valuable since derived from the cell by bin matrix?
• Would be useful if our heatmap viewer can scale

2. Quality control

• Potentially view in a table

3. Motifs

• In higlass: potentially nucleotide track or epilogos?

4. Variability scores for each motif

• Which transcription factors are enriched per cell

5. Quality control scores motifs
6. Probably more helpful to visualize cluster profiles rather than individual cell profiles

• Is there a clustering available in the snap output files already?
• Also would want to show variability within clusters

7. Can we summarize each cluster into a single profile (using the bin data for profiles)

• Link from cell sets to clusters (e.g. with colors)

[ngehlenborg]

Related to hubmapconsortium/portal-ui#1334:

• Stanford bulk ATAC-seq files (NA_summits.bed, NA_peaks.narrowPeak) can be visualized as a single, overlaid HiGlass track in Vitessce (if HiGlass doesn't support bed files, we can use Gosling)
• We need to decide which additional genome annotation tracks we should be displaying. At a minimum, we should have a track with gene locations for the correct genome build.
• We need to be able to figure out which genome build is being used.

@mruffalo How can we figure out which genome build was used to process these datasets?

[mruffalo]

@ngehlenborg What exactly do you mean by "figure out" -- what type of answer do you have in mind? Me answering in a comment to this GitHub issue? (GRCh38 with GENCODE v32 anntoations for all processed ATAC-seq datasets.) Storing a mapping of pipeline versions (commit hashes? tags? both?) to genome and annotation versions in this repository or somewhere else appropriate? Or a programmatic way to obtain the annotations for a derived data set, given the pipeline version that was used to produce that data set?

Something like this could be automated by examining a derived data set, obtaining the pipeline commit that produced that data set, and getting supplementary data from the appropriate Docker image:

$ docker run -it --rm hubmap/sc-atac-seq-grch38:1.2-bulk
root@2ff4069dd1db:/opt# ls -1 supplementary-data/
bwa-index
gencode.v32.annotation.bed
grch38.fasta.fai
hg38.blacklist.bed
hg38.promoters.bed

This would allow accessing the actual genome annotations in BED format -- does something like this seem useful enough to make more convenient?

[ngehlenborg]

Sorry, that wasn't very clear. I am wondering how we can figure out which genome build was used programmatically. We should probably have that for each pipeline through an API or a well-defined location in the CWL file?

I am not sure what is best, but I would rather not have to write code that checks file names on disk.

[ilan-gold]

I agree with @ngehlenborg - the way this would work ideally is that it would be somewhere that is eminently parse-able (say some sort of metadata.tsv or json file) so that the portal backend can pick it up and throw it in the config for Vitessce, which will then fetch the correct annotation for that genome. I think the CWL file is a good location too - the most important thing will be consistency at least within each assay, if not across assays, that need this sort of thing.

[ngehlenborg]

We need to agree on a location for the genome build for a given data set with the IEC and the CMU TC. Added to portal call agenda.

[mccalluc]

From the 1/21/2022 minutes:

Genome build info communicated in the output directories to be used with the index

• Need to know which reference genome to use and reference genome to display

MR is going to add this feature to the ATAC and RNA pipelines

• Vitessce will utilize the information and a .json file should be sufficient

cc @mruffalo : Please update here if that isn't correct.

[mccalluc]

Matt posted on hive-developers February 9:

{
    "genome": "grch38",
    "annotations": {
        "source": "GENCODE",
        "version": 35
    }
}

Nils responded:

Confirmed that this is sufficient. Matt Ruffalo, You can go ahead and get this out.

[mccalluc]

Ilan says:

This just kind of fell by the wayside. I’ll have a look again, I don’t remember where this was left.