Scope¶

RustScenic focuses on the regulatory-network compute path that benefits most from a small Rust-backed Python package: matrix-level inference, per-cell scoring, motif enrichment, topic modelling, enhancer-gene links and eRegulon assembly.

Designed For¶

• Local CPU runs on laptops and workstations.
• Python 3.10 to 3.13 environments.
• Researchers who want fewer moving parts than the legacy SCENIC stack.
• Benchmarked, reproducible workflows with commands and artefacts committed in the repository.

Current Boundary¶

• Motif ranking databases are external inputs because public databases can be hundreds of megabytes to tens of gigabytes.
• GRN edge rankings are independently implemented and can differ from arboreto at fine grain because RustScenic uses an independent histogram-tree builder; the early-stop monitor and fitted-tree distribution are validated separately.
• Topic modelling ships both Online VB and collapsed Gibbs paths. The Gibbs path is the stronger sparse scATAC option at larger topic counts.
• The million-cell benchmark measures RNA gene-network inference, not a complete spatial or RNA/chromatin workflow. Preparation memory is reported separately.
• Full workflow coverage from raw fragments plus external motif databases is in active validation.

Positioning¶

RustScenic combines gene-regulation analysis stages in a CPU-based Python package. Benchmarks show faster execution on the stated workloads; output agreement varies by stage. Reproducibility requires the same input, version, seed, thread count and settings.