msmu.pp.split_tmt

Split TMT channels in a MuData object into separate modalities based on a mapping.

Splitting relabels the C reporter channels into C x n_set distinct channel_set samples so that each biological sample is unambiguous. Because a given PSM is measured in only one set, the resulting PSM matrix is block-diagonal: every feature carries values in just its set's channels and is missing (NaN) in every other set's. A dense store would be O(n_set^2) -- both the feature count and the sample count scale with the number of sets, which is what makes many-plex studies exhaust memory -- so only the block diagonal (O(n_set)) is stored, as a SciPy sparse .X. Every obs label and stored value is identical to a fully materialised split; only the structurally-absent cross-set cells are left out.

Note: because those cells are not stored, mdata.mod["psm"].to_df() returns 0 (SciPy sparse convention) rather than NaN for them. Inspect the matrix with :func:msmu._core._blockdiag.dense_block, which restores absent cells as NaN.

Parameters:

Name	Type	Description	Default
mdata	MuData	The MuData object containing TMT data.	required
map	dict[str, str] | Series | DataFrame | None	A mapping of filenames to set names. If a DataFrame is provided, it should have two columns: the first for filenames and the second for set names. If None (the default), the map is derived from the attached SDRF (comment[data file] -> set_key, one set per file), which requires attach_sdrf first.	None
set_key	str	SDRF column naming each file's set/plex when deriving the map (map=None). Default comment[sample preparation batch]; name another per-file-constant column (e.g. a factor value[...]) when the SDRF encodes the set elsewhere.	'comment[sample preparation batch]'

Returns:

Type	Description
MuData	The MuData object with TMT channels split into channel_set samples.