Release notes¶

v2.0.0 (2026-05-11) — package renamed scsplice¶

Breaking change: The PyPI distribution name changes from splikit-py to scsplice. The canonical import alias is now scs (parallels sc / scv / sq).

# Before (splikit-py)
pip install splikit-py
import splikit as splk

# After (scsplice)
pip install scsplice
import scsplice as scs

Migration guide¶

v1.0 (splikit-py)	v2.0 (scsplice)
`import splikit as splk`	`import scsplice as scs`
`splk.io.read_starsolo(...)`	`scs.io.read_starsolo(...)`
`splk.tl.make_m2(adata)`	`scs.tl.make_m2(adata)`
`splk.pp.highly_variable_events(adata)`	`scs.pp.highly_variable_events(adata)`
`splk.tl.pseudo_correlation(adata, Z)`	`scs.tl.pseudo_correlation(adata, Z)`
`adata.uns["splikit"]`	`adata.uns["scsplice"]`
`pip install splikit-py`	`pip install scsplice`
`SPLIKIT_REAL_DATA_DIR` env var	`SCSPLICE_REAL_DATA_DIR` env var

`uns["splikit"]` compatibility shim¶

AnnData objects created with v1.0 carry uns["splikit"]. In v2.0, the canonical key is uns["scsplice"]. The shim in scsplice._core._validators.get_scsplice_ns() and setdefault_scsplice_ns() reads from uns["splikit"] with a FutureWarning and migrates the value to uns["scsplice"] automatically. The legacy key will be removed in v3.0.

To migrate an existing AnnData object explicitly:

import scsplice as scs
import warnings

# Load an old AnnData (will emit FutureWarning on first scsplice call)
with warnings.catch_warnings(record=True):
    scs.tl.make_m2(adata)  # shim fires, migrates uns key

# adata.uns["scsplice"] is now populated; adata.uns["splikit"] is removed

`SCSPLICE_REAL_DATA_DIR` env var¶

The SPLIKIT_REAL_DATA_DIR environment variable used by the test suite is renamed to SCSPLICE_REAL_DATA_DIR. The old name is still accepted with a FutureWarning for one release cycle; update your shell config.

v1.0.0 (2026-05-11)¶

Initial PyPI release as splikit-py.

Scope: six public functions across io, tl, and pp.

Cross-language parity: M2 bit-exact vs. R splikit; HVE deviance rtol=1e-10; pseudo-correlation rtol=1e-7. The full regression suite lives on the validation branch.

New in v1.0.0¶

scs.io.read_starsolo — Ingest STARsolo Solo.out/SJ/ for one or more samples into a single AnnData with M1 (inclusion counts) in layers["M1"] and LJV grouping in var["group_id"]. Supports spatial data via optional tissue_positions= parameter.

scs.io.read_starsolo_gene — Ingest Solo.out/Gene/ gene-expression counts into a standard cell × gene AnnData with raw UMI counts in X. Drop-in for scanpy.pp.normalize_total, scanpy.pp.highly_variable_genes, and scvi-tools. Supports tissue_positions= for spatial samples with full squidpy obsm["spatial"] and uns["spatial"] population.

scs.io.read_starsolo_velocyto — Ingest Solo.out/Velocyto/ spliced/unspliced/ambiguous layers into an AnnData compatible with scvelo. Handles both modern (split-file) and legacy (stacked matrix.mtx) STARsolo wire formats.

scs.tl.make_m2 — Build the exclusion matrix M2 from M1 and LJV grouping via C++ kernel with optional OpenMP parallelism. Output is bit-exact with R splikit.

scs.tl.pseudo_correlation — Per-event signed pseudo-R² (Cox-Snell or Nagelkerke) against an external predictor matrix via iteratively reweighted least squares (IRLS).

scs.pp.highly_variable_events — Select highly variable splicing events per library using binomial-deviance scoring.

scsplice.settings — Global settings object for configurable behavior (verbosity, I/O defaults).

MkDocs Material documentation site with tutorials, how-to guides, API reference, and conceptual explanations.

Known limitations¶

scs.pl (plotting) is a v2.1 placeholder. Use scanpy.pl, scvelo.pl, and squidpy for downstream visualization.
HVG (highly variable genes) and silhouette metrics are intentionally not ported — compose with scanpy.pp.highly_variable_genes and sklearn.metrics.silhouette_score instead.
GTF annotation and gene-level plotting are out of scope; use pyranges for GTF operations.
R-equivalence validation suite (cross-language regression tests, R fixtures, tolerance bands) lives on the validation branch, not main.