v0.4.4 - RNA004

Highlights

🎉 Full RNA004 Support with the WDX4 model, achieving 99.5% target accuracy at 95% yield. Features barcode-specific target accuracy confidence score filtering for optimal performance.

Major Improvements

New Detection Workflow

• Integrated ADAPTed v0.2.3 with CNN-based adapter detection
• Faster and more accurate boundary prediction for adapters
• Replaces previous LLR workflow as primary detection method for RNA004
• New dependency: PyTorch

Enhanced File Processing

• Streamlined output handling with direct barcode prediction streaming
• No intermediate files created
• Separate thread for read preloading with efficient I/O operations
• Controlled mini-batch processing to limit memory footprint

Improved User Experience

• Dynamic progress tracking using parallel pod5 read counting
• Simplified signal normalization with MAD-based outlier clipping
• Structured output directory organization:
  • predictions/: Demultiplexed read outputs
  • failed_reads/: Unsuccessful adapter detection attempts
  • Directory naming includes model, chemistry, version, and UUID

New Recovery Features

• Added retry command for processing failed reads
  • Uses alternative combined_detect_llr2 workflow
  • Complements existing continue functionality
• Streamlined command set:
  • demux: Primary demultiplexing from pod5
  • continue: Resume incomplete runs
  • retry: Recover failed reads

Performance Evaluation RNA004

[Performance metrics and visualizations included in attached figures]

• Precision and recall metrics for WDX4 and WDX10
• Confusion matrices demonstrating minimal cross-talk
• Calibration performance analysis
• Accuracy-yield trade-off curves

v0.4.3

[v0.4.3] - 2024-09-11

Added

• Added a --chemistry flag to the parser. This can be used to specify the latest default config TOML file.
• command.json file is now saved in the output folder and contains the command used to run WarpDemuX.
• Logging: process outputs are now logged to the warpdemux.log file and to stdout.
• WarpDemuX now supports continuing from a previous (incomplete) run using the continue <continue_from_path> subcommand.

Fixed

• setup.py: added a config_files folder to the package data.
• signal preload size is now explicitly updated based on the max_obs_trace value after initializing the config object.

Changed

• SigProgConfig loader functions now load the config in a two-step approach. First, they load the version&chemistry specific adapted config file, then they update the config with the chemistry-specific or user-provided config file. This allows to remove redundant parameters from the warpdemux config files.
  `
• Config names now include the translocation speed of the chemistry, ie: rna002_70bps@v0.4.3.
• The created output subdirectory path now contains the version of warpdemux used.
• Now depends on adapted v0.2.2.
• Event segmentation now relies on true peak detection in the t-scores, rather than sort-and-select. The num_events highest peaks in the calculated t-statistics, as detected with a minimal distance of min_obs are returned as changepoints.
• save_dwell_times in parser is False by default (True before).
• The output directory is now named after the version of WarpDemuX and a random UUID rather than the current date and time.

Removed

• The --create_subdir argument is removed. The output directory is now always created in the specified output folder.