Refactor/output direct to h5ad

[noamteyssier]

No description provided.

[gemini-code-assist]

Code Review

This pull request replaces the external Python-based pycyto convert step with native Rust writing of .h5ad files directly from cyto-ibu-count using the anndata and anndata-hdf5 crates, eliminating the need to stage through MTX files or spawn a Python subprocess. To prevent spurious file-locking issues under concurrent Rayon execution, subprocesses for cell-filter and geomux are now invoked with HDF5_USE_FILE_LOCKING=FALSE. The review feedback focuses on improving the robustness and performance of the new write_counts_h5ad function. Specifically, it is recommended to dynamically determine the number of observations (n_obs) from the populated obs_names vector to prevent potential mismatches, and to pass the features vector by value (Vec) instead of a slice to avoid cloning the entire list of feature strings.

[gemini-code-assist]

Using counts.get_num_barcodes() as n_obs before the loop can lead to a mismatch or failure if some barcodes are filtered out or if there is any discrepancy with the actual number of unique barcodes decoded in the loop. Since obs_names is populated dynamically during the loop, it is much safer and more robust to define n_obs as obs_names.len() after the loop. We can still use counts.get_num_barcodes() to pre-allocate the capacities of obs_names and bc_idx_map.

Suggested change

	let n_var = features.len();
	let n_obs = counts.get_num_barcodes();
	let nnz = counts.get_nnz();
	let mut obs_names = Vec::with_capacity(n_obs);
	let mut bc_idx_map = HashMap::with_capacity(n_obs);
	let n_var = features.len();
	let n_barcodes = counts.get_num_barcodes();
	let nnz = counts.get_nnz();
	let mut obs_names = Vec::with_capacity(n_barcodes);
	let mut bc_idx_map = HashMap::with_capacity(n_barcodes);

[gemini-code-assist]

Use obs_names.len() as n_obs to ensure the number of rows in the sparse matrix matches the length of obs_names exactly, preventing any potential out-of-bounds errors or mismatches.

Suggested change

	let coo = CooMatrix::try_from_triplets(n_obs, n_var, row_idx, col_idx, vals)
	.map_err(|e| anyhow::anyhow!("Unable to build sparse count matrix: {e}"))?;
	let n_obs = obs_names.len();
	let coo = CooMatrix::try_from_triplets(n_obs, n_var, row_idx, col_idx, vals)
	.map_err(|e| anyhow::anyhow!("Unable to build sparse count matrix: {e}"))?;

[gemini-code-assist]

To avoid cloning the entire vector of feature strings (which can contain tens of thousands of elements), we can pass features by value as Vec<String> instead of a slice &[String]. Since write_counts_h5ad is the final step in the counting process and consumes the features, passing by value is highly efficient and avoids unnecessary heap allocations.

Suggested change

	pub fn write_counts_h5ad<P: AsRef<Path>>(
	path: P,
	counts: &BarcodeIndexCounts,
	features: &[String],
	header: Header,
	suffix: Option<&str>,
	) -> Result<()> {
	pub fn write_counts_h5ad<P: AsRef<Path>>(
	path: P,
	counts: &BarcodeIndexCounts,
	features: Vec<String>,
	header: Header,
	suffix: Option<&str>,
	) -> Result<()> {

[gemini-code-assist]

Since features is now owned by the function, we can pass it directly to DataFrameIndex::from without calling .to_vec(), completely eliminating the need to clone the feature strings.

Suggested change

	adata.set_var_names(DataFrameIndex::from(features.to_vec()))?;
	adata.set_var_names(DataFrameIndex::from(features))?;

[gemini-code-assist]

Pass features by value to write_counts_h5ad instead of taking a slice, to avoid cloning the feature strings inside the function.

            features
                .expect("Must provide a feature file to write h5ad"),