Repository navigation
convert: CellBender remove-background output, with its latents - #51
Conversation
Fixes #29. A CellBender output.h5 or output_filtered.h5, 0.2 or later, is recognised by its latents and read as a 10x matrix plus: - per-droplet latents as obs columns, NaN for droplets CellBender did not analyse, and obs["cellbender_analyzed"]; - the gene-expression encoding in obsm, the ambient profile in var, and everything else in uns["cellbender"]; under the names CellBender's own loaders give, so its notebooks keep working. Which latent belongs to which barcode follows CellBender's own rule: 0.3+ filtered files still store every analysed droplet's index. --min-cell-probability keeps likely cells. --raw adds the Cell Ranger counts CellBender ran on as layers["cellranger"] (and X again as layers["cellbender"]), matched by barcode: straight across for the full output, through the on-disk buckets for the filtered one, whose cells CellBender orders by count. That path is now core/csr_build.write_runs, shared with the MTX import; fan_out_sparse_matrix can name each target's matrix, so X and layers["cellbender"] come from one read. An .h5 output is still a 10x matrix: outputs pick the first format with that extension that can be written, and --to a read-only format says so. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.
An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Test Results (py3.12)1 441 tests 1 441 ✅ 3m 25s ⏱️ Results for commit 3db9c67. ♻️ This comment has been updated with latest results. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 533ac38cfc
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 address that feedback".
Test Results (py3.13)1 441 tests 1 441 ✅ 3m 23s ⏱️ Results for commit 3db9c67. ♻️ This comment has been updated with latest results. |
From review on #51: - The embedding was read whole before the block loop. Latents are now read in slices and sorted into output blocks through scratch, since fancy indexing scattered rows is ~500x slower than a slice. - --raw matched barcodes through a dict of the whole column. It now picks the output's barcodes out of contiguous reads and looks raw blocks up in a sorted numpy array. - Raw-layer bucket sizes were estimated from feature widths, which rounds to zero for a narrow filter. A counting pass gives exact sizes. Also adds var["cellbender_analyzed"] from features_analyzed_inds, as CellBender's anndata_from_h5 does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fixes #29.
adata convert output.h5 -o out.h5adnow recognises CellBender'sremove-backgroundoutput (0.2, 0.3 and 0.4 layouts, full and filtered) by its latents, whatever the file is called. X is the denoised counts, and the latents are kept under the names CellBender's own loaders use:cell_probability,cell_size,background_fraction,droplet_efficiency;latent_*for 0.2), NaN for droplets CellBender did not analyse, plusobs["cellbender_analyzed"];obsm["gene_expression_encoding"](latent_gene_encodingfor 0.2), written in row blocks;var["ambient_expression"];uns["cellbender"].New flags:
--min-cell-probability Pkeeps droplets withcell_probability >= P, in file order.--raw raw_feature_bc_matrix.h5addslayers["cellranger"](the raw counts of the same barcodes) andlayers["cellbender"](a copy of X), asload_anndata_from_input_and_outputdoes. Filtered files list cells by UMI count rather than file order, so their raw counts go through the on-disk bucket sort; that code moves fromformats/mtx.pyintocore/csr_build.write_runsand is shared.--from 10x-h5still reads a CellBender file as a plain 10x matrix. Registering a read-only.h5format also changed how output extensions resolve: an extension now picks the first format that can be written, and--to cellbenderis refused with "can be read but not written".Tests: 21 in
tests/test_cellbender.py(every layout, h5ad and Zarr, each latent at its barcode,--rawwith filters and splits, refusals that write nothing), a scanpyread_10x_h5comparison, and perf guards for linear work and block-wise obsm writes.🤖 Generated with Claude Code