SPLIT v0.3.0 – Remove residual contamination

Residual contamination removal

Standard SPLIT purifies spatial transcriptomics profiles by proportionally
redistributing signal between assigned cell types. Some genes, however, are
not specific to any of the assigned cell types — they represent contamination
from cell types entirely absent from the local deconvolution model and cannot
be resolved by proportional redistribution alone.

This release introduces DO_remove_residual_contamination = TRUE in purify(),
which explicitly zeros out these non-specific genes using the reference expression
profiles. The specificity threshold is automatically derived from the reference
distribution, requiring no manual tuning. On a public Xenium breast cancer
dataset, this removes an additional ~2–5% of counts on top of standard SPLIT,
with a measurable improvement in marker gene specificity.

See the dedicated vignette for a full comparison.

Other changes

• Remove epsilon from the purification formula — genes with no reference
  support in the primary cell type are now explicitly zeroed rather than
  receiving a small residual value, giving cleaner output across all modes.
• New vignette Residual_Contamination_Removal.Rmd.
• Updated Run_RCTD_and_SPLIT_on_Xenium.Rmd with notes on the
  annotation-method agnostic interface and py-RCTD.

Backward compatibility

All changes are fully backward compatible. DO_remove_residual_contamination
defaults to FALSE — existing purify() calls are unaffected.

---

Full changelog: #27