Add Chai-1

[jscgh]

Description of feature

Add Chai-1 to the pipeline.

Chai-1 is a multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of benchmarks. Chai-1 enables unified prediction of proteins, small molecules, DNA, RNA, glycosylations, and more.

Paper

A proof of concept nextflow implementation already exists as nf-chai.

Separately we have a branch on our repo which was created in response to a request to test Chai-1. As of this post, it is capable of running in a barebones state.

[jscgh]

The version available on our branch is working but writing out the rest of the boilerplate files is not currently a priority.

If anyone is interested in taking over I have listed the pending tasks below:

Required

• [x] Needs subworkflows/local/prepare_chai1_dbs.nf
• [x] Needs conf/test_chai1.config
• [x] Needs conf/modules_chai1.config
• [ ] Complete the steps on the PR checklist

PR checklist

• [ ] Usage Documentation in docs/usage.md is updated.
• [ ] Output Documentation in docs/output.md is updated.
• [ ] CHANGELOG.md is updated.
• [ ] README.md is updated (including new tool citations and authors/contributors).
• [x] Make sure your code lints (nf-core lint).
• [x] Ensure the test suite passes (nextflow run . -profile test,docker --outdir <OUTDIR>).
• [x] Check for unexpected warnings in debug mode (nextflow run . -profile debug,test,docker --outdir <OUTDIR>).

Optional

• [ ] ARM64 incompatible currently due to Gemmi package (only macos arm64, no aarch64)

[FloWuenne]

@jscgh I would love to work on this during the hackathon since we have some GPU availabilities for the duration of the hackathon that I can use to make this work!

[jscgh]

I have updated the dockerfile for the Chai-1 image for a ~25% reduction in size. It is available on our fork here.

This won't change functionality at all and should slot right in with whatever other work has been done.

[FloWuenne]

Amazing work @jscgh ! I have to admit I didn't get enough time to finalize a module for this, but have it on my radar to continue this work in the coming weeks. Will ping here with any updates. Smaller containers are definitely a great start!

[jscgh]

I've added prepare_chai1_dbs.nf; however, it's not able to download the models yet.

It needs to iterate over components = ["feature_embedding", "token_embedder", "trunk", "diffusion_module", "confidence_head"] to download chai1_models_link"/${component}.pt"

I wasn't able to get that working in nextflow.