Add HDF5-backed CytoTalk cached runner script

[pooriass]

Summary

• add an R script that builds a file-backed mutual information cache for CytoTalk runs
• implement a streamed ARACNE DPI pruning routine to avoid loading full MI matrices into RAM
• provide a cached runner helper and example invocation for a cell-type pair

Testing

• not run (not applicable)

---

https://chatgpt.com/codex/tasks/task_b_68f2390e87dc8320a8e5e3af4d5d8d3b

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting

• @codex fix this CI failure
• @codex address that feedback

[chatgpt-codex-connector]

Prevent neighbor-loop from indexing beyond bounds

The streamed ARACNE routine loops neighbor pairs with for (a in seq_along(idx_nz)) and for (b in (a + 1L):length(idx_nz)). When a reaches the last neighbor, (a + 1L):length(idx_nz) evaluates to a decreasing sequence containing values larger than length(idx_nz) (e.g. 3:2), so idx_nz[b] becomes NA and the subsequent h5read(..., index = list(j, k)) call uses an NA index and aborts. This will throw an error for any node that keeps two or more neighbors unless the last neighbor was already pruned, making the streamed mode unusable. Guard the inner loop to run only when a < length(idx_nz) (e.g. iterate seq_len(length(idx_nz) - 1L) or skip when there is no next neighbor).

Useful? React with 👍 / 👎.