The "some day" List¶

Stuff that I'd like to get to at some point that isn't important enough to prioritize now.

Development¶

Cram convenience/plotting functions into a R package for ease of maintenance/sharing.
Find a way to summarize lots of GO terms into a human interpretable output that is both succinct and doesn't suck.
- Fine tune an LLM and use RAG with the term descriptions?
- Preferably from within R without relying on external setup to run models.
Write up CRISPRball for JOSS or such & submit.
- Get public version live on DMZ VM.
Make IBET more performant.
- Avoid complete re-plotting when adding labels.
- Add option to add labels automatically on highlighted gene(sets).
Make the nf-core Cut & Run pipeline script for fragment length histograms more memory efficient by changing from using np.array for each iteration to a list and concatenating the list objects into an array at the end.
- This sucker takes like 500+ GB of memory for large datasets unnecessarily.

Typical params.yaml files for nf-core pipelines pointing to our reference data rather than having that crammed into configs.
Template notebooks for ATAC-seq.
Template notebooks for ChIP-seq/CnR.
Set up nf-core rnafusion for easy running on RNA-seq data.

Get genesets from: https://www.science.org/doi/10.1126/science.add7046
Make combined reference genomes for nf-core pipelines, e.g. mm10 + hg38. GTFs and FASTAs, with clear species indicators.
- At least for RNA-seq, this takes advantage of the fractional counts based on probability (EM) from salmon for ambiguous reads.
- Have clear scripts for creation.