Skip to content

APPLY_ALL_SUMMARY.md

Latest commit

 

History

History
107 lines (79 loc) · 2.77 KB

APPLY_ALL_SUMMARY.md

File metadata and controls

107 lines (79 loc) · 2.77 KB

--apply_all Flag: Quick Reference

What It Does

Extends feature calling to all barcodes in the matrix, not just the filtered list.

  • Trains on filtered barcodes (high quality)
  • Applies learned model to all barcodes passing quality filters
  • Rescues additional cells that would otherwise be excluded

Usage

# Basic usage
call_features \
  --mtx-dir /path/to/matrix \
  --cell-list filtered.tsv \
  --out-prefix results/rescue \
  --apply-all \
  [other flags as normal]

Status

✅ Production Ready

  • Regression tested and validated
  • Filtered barcodes: identical assignments
  • Memory/runtime overhead acceptable
  • All three modes supported

Key Features

  • Backward compatible: No impact when flag not used
  • Conservative: Same quality filters (M_min) apply to all
  • Simple output: All results in same files (filtered + unfiltered)
  • Tested: Comprehensive regression test suite included

Mode-Specific Behavior

Mode Implementation Status
simple-assign Applies min_count/min_ratio ✅ Fully validated
FLEX Full binomial test + decision rules ✅ Production ready
EM Stored fit parameters + heuristic ✅ Validated*

*EM uses calibrated heuristic approximation; recommend monitoring on production data

Performance

  • Memory: +200-500 MB for 1.6M barcode matrices
  • Runtime: +10-30% (depends on unfiltered barcodes passing M_min)
  • Scales linearly with matrix size

Documentation

  • User guide: ../README.md section 2.5
  • Technical details: Technical.md section 7
  • Testing results: TESTING_COMPLETE.md
  • Change log: ../CHANGELOG.md

Testing

Run regression tests:

cd /mnt/pikachu/call_features
./scripts/test_apply_all_regression.sh

Expected: All tests pass ✅

Quick Start Examples

Simple-assign (most validated)

call_features --mtx-dir matrix/ --cell-list filtered.tsv \
  --out-prefix rescue/ --apply-all \
  --simple-assign --min-count 3 --min-ratio 2.0

FLEX (statistical rigor)

call_features --mtx-dir matrix/ --cell-list filtered.tsv \
  --out-prefix rescue/ --apply-all \
  --tau 0.8 --delta 0.4 --gamma 0.9

EM (complex screens)

call_features --mtx-dir matrix/ --cell-list filtered.tsv \
  --out-prefix rescue/ --apply-all \
  --use-em --k-min 4 --tau-pos 0.95

Monitoring Recommendations

After deployment:

  1. Compare QC stats baseline vs --apply_all
  2. Check rescued barcode quality (UMI distributions)
  3. Verify memory usage within expectations
  4. Monitor doublet rates in rescued population

Support

  • Implementation: ../src/call_features.c lines ~703-1511
  • Tests: ../scripts/test_apply_all_regression.sh
  • Issues: Check TESTING_COMPLETE.md for known limitations