feat: scaffolding, caching, EGFR

[tristan-f-r]

We bundle EGFR along with the rest of the caching infrastructure. Notes:

• All motivation for the caching system lives under cache/README.md.
• We removed pra.yaml for now, as the only PRAs are the synthetic data and the ResponseNet data, and soon the DepMap data.
• The CONTRIBUTING.md file is in Changes to CONTRIBUTING guide #57.
• directory.py contains unnecessary files from other datasets that were deemed universal.

[ntalluri]

I did a light review of the PR; did not look to hard at the code itself yet. I mostly was gathering ideas on what was happening from the READMEs.

[tristan-f-r]

I implemented the approach to versioning above, which I'll document, but effectively:

• If a person wants to just use 'STRING', they say 'STRING/latest' and it will point to STRING v12
• If a dataset wants to use a specific version of STRING, they say 'STRING/v12'

This is naively implemented by allowing strings in directory that point to other keys.

[ntalluri]

Here is another review of this PR.

Main issues:

1. Why are we still erroring when pinned doesn't = cached? I thought we agreed that pinned != cached will be a warning at the last meeting.

2. What happens when this error above occurs, how is a user supposed to fix it to allow them to download the data and get rid of the error?

3. having data in directory.py and the dataset specific snakefiles doesn't make sense (and also having the cached files in two places is also confusing). We need to decide to either put everything in directory.py or in the dataset specific snakefiles and have the contributing guide reflect this.

• How is a user supposed to know when their data is supposed to be used in multiple other datasets? That's a very large expectation for them to know that.

4. Why is there no config for the egfr dataset collection?

[ntalluri]

Also not sure why the workflow is breaking.

[tristan-f-r]

As for (1) and (2), we did decide this! I don't know if I didn't push it or what, but I did completely intend to make that not an error. I forgot to remove the move/copy flag 👍 this has been removed in the non-styling commit after this comment.

[tristan-f-r]

My two notes:

• How important is it that the 10 value stays 10? (e.g. does 1000 work just as well?) The methodology described in the paper says that it just needs to be a value greater than every other prize.
• Unless (*) it is the case that we plan to have different algorithm parameter inputs for different dataset collections, I disagree with the methodology that the paper describes for a config per dataset collection, where I would instead believe it to be straightforward to continue as is. Otherwise, unless (*) is true, we directly contradict one of SPRAS's usability goals of running several datasets on several algorithms.

[ntalluri]

Based on meeting:

1. update to have datasets fetch configs either in directory.py or in a snakemake file that is dataset specific (one or the other not both)

• this idea will be updated in the contributing guide as well

2. keeping the value of 10 for the prizes for egfr

• hard to justify but that's just the plan for now

3. we will have 4 separate dataset collection configs (we can also keep the test-config too if needed).