Changes in version 2.2.0 (2026-01-23)                  

Code Quality Improvements

API Refinements

  - Renamed construct_flux_boundaries_fast → construct_flux_bounds
    (cleaner API)
  - Renamed internal function parameters for clarity:
      - LB_rep → lb_base, neg_one_vec → lb_tail, etc.
  - Renamed use_parallel → parallel, num_cores → n_cores, use_rcpp →
    use_cpp

Code Cleanup

  - Removed redundant comments (e.g., "优化", "修复", "BUG FIX")
  - Standardized internal function naming with . prefix
  - Improved roxygen2 documentation with detailed parameter descriptions
  - Consistent English documentation throughout

Bug Fixes

  - Fixed calculate_multi_comp() logic: unique(com) == TRUE → all(com)
  - Fixed compute_flux() indexing: proper handling of reversibility
    indices
  - Fixed biomass reaction bounds: biomass now uses original UB (1000)
    instead of MRAS
  - Full Seurat v4/v5 compatibility via GetAssayData()

Cross-Platform Stability

  - PSOCK clusters for Windows/macOS/Linux compatibility
  - Guaranteed cluster cleanup with on.exit()
  - Auto-detection of physical CPU cores (capped at 8)
  - Pre-compiled C++ functions (no runtime compilation)

                 Changes in version 2.1.1 (2026-01-20)                  

Bug Fixes

  - Fixed boolean vector handling in reaction score calculation
  - Fixed integer index negation bug in flux computation
  - Added Seurat v5 compatibility layer

                 Changes in version 2.1.0 (2025-01-25)                  

Major Performance Improvements (10-20x faster)

Parallel Computing (4-8x speedup)

  - Multi-core parallel processing using parallel, foreach, and
    doParallel
  - Auto-detects available CPU cores (default: all cores - 1)
  - New parameters: use_parallel = TRUE, num_cores = NULL

Rcpp Optimization (1.05-1.1x speedup)

  - C++ implementation for boundary construction
  - 10-50x faster than R version for boundary calculations
  - Automatic fallback to R if Rcpp unavailable
  - New parameter: use_rcpp = TRUE

Combined Performance

  - Total speedup: 10-20x for typical analyses
  - Example (100 bootstraps × 5 cell types):
      - Before: ~17 minutes
      - After: ~1-2 minutes (with 8-core CPU)

Backward Compatibility

  - 100% compatible - all existing code works without changes
  - Default behavior: optimizations enabled automatically
  - Can disable: compute_sc_flux(..., use_parallel = FALSE)

Bug Fixes

  - Fixed critical bug in calculate_avg_exp() bootstrap sampling
    (v2.0.1)
  - Fixed METAFLUX::: namespace references
  - Fixed Rcpp function export in parallel workers

New Dependencies

  - parallel, foreach, doParallel - for parallel computing
  - Matrix - sparse matrix operations
  - Rcpp (optional) - C++ optimization

                        Changes in version 1.0.0                        

Initial Release

This is the first release of METAFLUX, a performance-optimized
implementation of METAFlux (Huang et al., Nature Communications, 2023).

Performance Improvements

Computational Speed

  - Bulk RNA-seq analysis: 30-60% faster
  - Single-cell RNA-seq analysis: 45-70% faster
  - Benchmark results:
      - 100 samples: 3.6 min → 2.0 min (44% improvement)
      - 500 samples: 18.3 min → 10.1 min (45% improvement)
      - 5K cells: 8.5 min → 4.2 min (51% improvement)
      - 10K cells: 26.7 min → 13.2 min (51% improvement)

Memory Efficiency

  - 30-50% reduction in peak memory usage
  - Optimized sparse matrix operations
  - Reduced memory allocation overhead

Technical Enhancements

Algorithmic Optimizations

  - Vectorized matrix operations replacing iterative calculations
  - Sparse matrix implementation using Matrix package
  - Pre-allocation of data structures to eliminate dynamic growth
  - Pre-computation of invariant values outside loops
  - Optimized bootstrap sampling for single-cell analysis

Code Quality

  - All external functions explicitly namespaced (package::function())
  - Comprehensive roxygen2 documentation
  - Type-safe operations with explicit data types
  - Consistent code style following tidyverse guidelines
  - Passes styler::style_pkg() without errors

Validation

Numerical Accuracy

  - All results numerically identical to original METAFlux
  - Difference < 1e-10 (floating-point precision limit)
  - Validated across:
      - NCI-60 bulk RNA-seq benchmark
      - Multiple single-cell datasets
      - Edge cases and boundary conditions

Test Coverage

  - Comprehensive validation test suite included
  - Unit tests for all core functions
  - Integration tests for complete workflows
  - Performance benchmarking utilities

API Compatibility

  - 100% backward compatible with original METAFlux
  - Identical function signatures and parameter names
  - Same input/output data structures
  - Drop-in replacement requiring no code changes

Dependencies

New Dependencies

  - Matrix (>= 1.3.0) - Required for sparse matrix optimization

Updated Requirements

  - R (>= 3.6.0, 4.0.0+ recommended)
  - osqp (>= 0.6.0)
  - dplyr (>= 1.0.0)
  - Seurat (>= 4.0.0) - for single-cell analysis
  - stringr (>= 1.4.0)
  - stringi (>= 1.5.0)

Documentation

  - Comprehensive README with installation and usage examples
  - Detailed optimization guide explaining technical improvements
  - Quick reference comparing original and optimized versions
  - Complete API reference documentation
  - Validation test suite with usage examples

Bug Fixes from Original

  - Fixed: styler syntax error in data.R documentation
  - Fixed: Potential memory leak in bootstrap sampling
  - Fixed: Edge case handling in MRAS calculation

Known Limitations

  - Requires Matrix package (new dependency)
  - Minimal floating-point differences (< 1e-10) may occur due to
    operation reordering
  - Performance gains most pronounced with larger datasets (>100 samples
    or >5K cells)

Migration from METAFlux v1

METAFLUX is a drop-in replacement:

# Simply change installation source
# Original:
# devtools::install_github('KChen-lab/METAFlux')

# METAFLUX:
devtools::install_github('Zaoqu-Liu/METAFLUX')

# No code changes needed
library(METAFLUX)  # Same API as original

Acknowledgments

This optimized implementation builds upon the excellent work of:

  - Yuefan Huang (Original algorithm and implementation)
  - Kyle Tsai (Development and testing)
  - Ken Chen Lab at MD Anderson Cancer Center