Highlights
- Adaptive Leiden graph resolution: Resolution parameter is now automatically computed based on data sparsity, eliminating manual tuning for different datasets.
- Pruning confidence criterion for CV selection: New
pruning_confidencestrategy detects and prevents over-pruning, ensuring biologically valid tree selection across diverse data qualities. - Cleaner output structure: Redundant files are no longer repeatedly saved; output directories are now more organized and easier to navigate.
What's New
Core Algorithm
-
Adaptive resolution for Leiden clustering (
scaffold_builder.py):- Resolution is now dynamically calculated from data quality metrics (NA ratio, rare mutation ratio, effective coverage)
- High-quality data → lower resolution; sparse/noisy data → higher resolution
- Eliminates the need for manual resolution tuning across different samples
-
Pruning confidence criterion for CV selection (
run_phylosilid_fullTree_scRNA.py):- New selection strategy:
pruning_confidence(now the default) - Computes
pruning_ratio = (Ω_pre-QC - Ω_final) / Ω_final - Confident pruning if
pruning_ratio < 10.0(the "One-Tenth Rule") - If confident trees exist: select the one with lowest
Ω_final - If no confident trees: fallback to lowest
Ω_pre-QC - Prevents over-pruning that previously caused loss of true biological signal
- New selection strategy:
Output Structure
- Cleaner output organization:
- Reduced redundant file saving across CV iterations
- Better directory structure for easier result navigation
- Consolidated log files with clear CV-specific separation
Bug Fixes
- None in this release (algorithm improvements and code quality enhancements only).
Breaking Changes
- CV selection strategy changed: The default selection criterion is now
pruning_confidence. Users who relied on the oldomega_final-only selection should review their results. - To revert to the old behavior, use
--selection_criterion omega_final(see documentation).
Performance
- Adaptive resolution reduces the need for manual parameter tuning
- Output cleanup reduces disk usage across multiple CV runs
Notes
- This release is recommended for all users as it significantly improves tree selection robustness.
- Fully backward compatible with v3.3.x workflows; existing command-line interfaces remain unchanged.
- For detailed usage, see the docs site: https://github.com/douyinlab/PhyloSOLID/