rsfgseapy 0.4.0

rsfgsea 0.4.0 — Decor and native Blitz GSEA

on Python PyPI

Highlights

Version 0.4.0 introduces three first-class GSEA tracks across Rust, CLI, Python, and R:

• redundancy-aware Decor GSEA
• classic fgsea-compatible execution
• native Blitz GSEA compatibility

Classic fgsea remains the default; Decor and Blitz are opt-in.

Decor GSEA

• Introduces expression-aware pathway scoring that downweights genes with redundant within-pathway correlation.
• Supports fixed-permutation simple execution and pathway-specific multilevel refinement.
• Adds reusable expression-derived redundancy caches.
• Includes calibrated sensitive, balanced, specific, and strict presets.
• Adds a 0–100 stringency control and an optional adaptive tail-reliability pass.

Native Blitz mode

• Adds a native Rust implementation targeting blitzgsea 1.3.54.
• Locks the complete Python reference environment for reproducibility.
• Reproduces SciPy 1.18's numerically stable de Boor-weight interpolation.
• Adds reusable in-process null-model caching and deterministic parallel execution.
• Achieves floating-point-scale agreement with the locked Blitz reference on the large 5,324-pathway audit.

Classic fgsea parity

On the same 63,904-rank, 5,324-result audit:

• ES and NES matched exactly for every pathway.
• Maximum p-value difference was 8.3e-17.
• Maximum adjusted-p-value difference was 2.8e-16.
• Leading-edge sets and finite-value classifications matched.

Additional improvements

• Adds machine-readable precision evidence and reproducible reference generators.
• Writes floating-point TSV values using round-trip-safe representations.
• Improves WSL2 GPU adapter selection and diagnostic guidance.
• Updates the project to Rust 1.97.1 and modernizes dependencies and release automation.
• Strengthens Linux, macOS, and Windows CI for Rust, Python, and R packages.

Full changelog: v0.3.4...v0.4.0