alphadia-search-rs 0.2.0

A next-generation search backend for AlphaDIA.

alphadia-search-rs

High-performance alphaDIA backend.

Notes for users

This repository contains the high-performance backend for alphaDIA. This code should to used as part of alphaDIA.

Development Setup

Prerequisites

• Rust 1.88.0
• Python 3.11+

Quick Start

1. Clone and enter the repository:

   git clone <repository-url>
   cd alphadia-search-rs
   

2. Set up pre-commit hooks (recommended):

   # Install pre-commit
   pip install pre-commit
   # or: conda install -c conda-forge pre-commit
   # or: brew install pre-commit
   
   # Install the git hook scripts
   pre-commit install
   

3. Install Python dependencies:

   conda activate alphadia-search-rs  # or create environment if it doesn't exist
   pip install maturin
   

4. Build the Rust extension:

   maturin develop --release
   

Omit the --release extension for a developer build.

5. Run tests:

   cargo test                    # Rust tests
   python ./scripts/test_search.py  # Python integration test
   

Testing

Integration Test

The scripts/test_search.py script provides a comprehensive integration test.

# Run the integration test
python ./scripts/test_search.py --path ./test_data

The script will automatically:

1. Use existing test data in ./test_data if available (using a temporary directory if --path not specified).
2. Otherwise download required files:
   • spectrum_df.parquet - Mass spectrometry spectra data
   • peak_df.parquet - Peak detection results
   • precursor_df.parquet - Precursor ion information
   • fragment_df.parquet - Fragment ion data

Expected output:

• Processing speed: ~200k+ precursors per second
• Results: ~11M candidates found

Scripts

The scripts/ directory contains analysis pipelines for processing DIA-MS data:

Key Scripts

1. Candidate Selection: Takes a calibrated speclib as an input and an AlphaRaw hdf. Performs candidate selection and saves the candidates.

   python scripts/candidate_selection.py --ms_data_path data.hdf --spec_lib_path lib.hdf --output_folder ./output
   

2. Candidate Scoring: Performs scoring following selection. Takes input from previous step and save precursor at 1% FDR.

   python scripts/candidate_scoring.py --ms_data_path data.hdf --spec_lib_path lib.hdf --candidates_path candidates.parquet --fdr --quantify
   

   • option to perform quantification with --quantify
   • option to perform FDR adn filter @1% with --fdr
   • option to add diagnosis plot for all features with --diagnosis

CLI Benchmarking

Score Benchmark Tool

The score-benchmark CLI tool benchmarks multiple implementations of axis_log_dot_product to compare performance and verify numerical accuracy.

# Run the benchmark
cargo run --bin score-benchmark

Troubleshooting

Library Loading Error on macOS: If you encounter the error dyld[xxxxx]: Library not loaded: @rpath/libpython3.11.dylib when running cargo test, set the library path:

Mac:

export DYLD_LIBRARY_PATH=$(realpath $(which python)/../../lib)
cargo test

Linux:

export LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH
cargo test

Development Workflow

Code Quality Standards

This project enforces strict code quality standards via automated tooling:

• Formatting: All code must be formatted with rustfmt
• Linting: All code must pass clippy with no warnings
• Consistency: Same toolchain used locally and in CI (Rust 1.88.0)

Pre-Commit Hooks

We use the pre-commit framework for automated code quality checks:

# Install pre-commit (one-time setup)
pip install pre-commit

# Install hooks (one-time setup)
pre-commit install

Manual Code Quality Checks

You can run the same checks manually:

# Format code
cargo fmt

# Check formatting (without modifying files)
cargo fmt --all -- --check

# Run linting
cargo clippy -- -D warnings

# Run all pre-commit hooks manually
pre-commit run --all-files