PIASO v1.1.0

Highlights

Faster gene set scoring via a new Rust backend, plus faster multi-batch GDR via ThreadPool parallelism. Backward-compatible.

What's New

Rust-Accelerated Scoring

• New score_complete() in Rust (PyO3): fuses control gene sampling, sparse matrix multiplication, and column-wise reduction into a single pass
• Releases the GIL during computation, enabling true thread-level parallelism
• LCG PRNG for deterministic control gene sampling (same seed per gene set)
• Automatically used when the Rust extension is available; falls back to Python otherwise

ThreadPoolExecutor for Multi-Batch Scoring

• calculateScoreParallel_multiBatch now processes batches concurrently using ThreadPoolExecutor (replaces sequential loop)
• New n_concurrent_batches parameter in runGDRParallel() for explicit control
• Auto-parallelism via _determine_parallelism(): balances inter-batch concurrency vs per-batch threads

Vectorized kNN Self-Loop Removal

• _precompute_stats() kNN cleanup rewritten with numpy gather+shift
• Eliminates Python list comprehension over all genes

Precomputed kNN

• New precomputed_knn parameter in score() — pass pre-built KDTree + indices to avoid redundant queries when scoring multiple gene sets on the same data

Build System

• Switched to maturin build backend for native Rust compilation during pip install
• Upgraded to PyO3 0.24 + numpy 0.24