deephalo 1.0.0

DeepHalo: A deep learning-integrated workflow for high-throughput discovery of halogenated metabolites from HRMS data.

DeepHalo

A deep learning-integrated workflow for high-throughput discovery of halogenated metabolites from HRMS data.

Core Features

1. Halogen Prediction

• Element Prediction Model (EPM)
  • Dual-branch Isotope Neural Network (IsoNN) architecture
  • High accuracy Cl/Br detection (>99.9% precision based on benchmark results)
  • Wide mass range coverage (50-2000 Da)
  • Robust interference resistance to B/Se/Fe/dehydro isomers

2. Isotope Pattern Validation

• Dual Validation System
  • Mass Dimension: Statistical rule-based correction.
  • Intensity Dimension: Autoencoder-based Anomaly Detection Model (ADM).

3. Multi-Level Halogen Confidence Scoring (H-score)

• Dual levels
  • Prediction based on centroid-level isotope patterns
  • Prediction based on Scan-level isotope patterns
  • H-score integration for comprehensive assessment on the above both levels

3. Enhanced Dereplication

• Dual-Strategy Approach
  • MS1-Based Dereplication Using Custom Database Matching
    • Exact mass analysis
    • Halogen presence verification
    • Isotope intensity similarity scoring
  • MS2-Based Dereplication by Integrating GNPS
    • MS2 molecular networking
    • Halogenated compound annotation
    • GraphML file enhancement

---

Technical Advantages

• High Throughput

  • end-to-end automated analysis
  • Batch processing of unlimited LC-MS/MS datasets
  • Rapid processing (several to dozens of seconds per sample) on standard hardware (Core i9, 16GB RAM)

• High Accuracy

  • 98.3% precision in halogen detection across simulated and experimental LC-MS datasets.

  • Comprehensively validation across both simulated and experimental LC-MS datasets

• Comprehensive Integration

  • Input: Supports .mzML format
  • Output: Cytoscape-compatible network files
  • Seamless integration with GNPS molecular networking

• Enhanced Dereplication

  • Embeds halogen prediction results into GNPS output GraphML files
  • Significantly higher dereplicaton rate compared to molecular networking alone

---

Target Applications

• Natural product discovery
• Halogenated metabolite annotation

---

Key Differentiators

1. Deep leaning-based halogen prediction resistance to Fe/dehydro isomers
2. First Isotope Pattern Validation strategies specific for halogenated molecules
3. hierarchical halogen scoring system (H-score)
4. Comprehensive dereplication workflow
5. Enhanced GNPS molecular networking

---

For methodology details and validation datasets, see Methods.

Where to get it？

The source code is hosted on GitHub at: https://github.com/xieyying/DeepHalo

Binary installers of DeepHalo are available at the Python Package Index (PyPI).

Dependencies

• pandas == 2.0.3
• numpy == 1.22.0
• molmass == 2023.8.30
• scikit-learn == 1.3.1
• tensorflow == 2.10.1
• keras == 2.10.0
• keras_tuner == 1.4.6
• matplotlib == 3.8.0
• pyopenms == 3.1.0
• scipy == 1.11.4
• tomli == 2.0.1
• tomli-w == 1.0.0
• importlib_resources == 6.4.0
• mzml2gnps == 1.0.3
• networkx == 3.4.2
• typer == 0.15.1

Installation

Note
Python 3.10 is required. Verify your Python version with:

 python --version

Install from PyPI

pip install DeepHalo

Install from Local Wheel

pip install path/to/DeepHalo-xxx.whl

Install from Source

git clone https://github.com/xieyying/DeepHalo.git
cd DeepHalo
pip install -e .

Quickstart

High-throughput Detection of Halogenated Compounds

halo detect -i /path/to/mzml_files -o /path/to/output_directory -ms2

Dereplication

halo dereplicate -o /path/to/output_directory -g /path/to/GNPS_results -ud /path/to/custom_database.csv

Full Usage Guide

Get help

halo --help                 # Show all commands
halo detect --help    # Detailed parameters for the subcommand 'detect'
halo dereplicate --help  # Detailed parameters for the subcommand 'dereplicate'

Main Functions

• Analyze mzML file:

  halo detect -i <input_path> -o <project_path> [-c <config_file>] [-b <blank_samples_dir>] [-ob] [-ms2]
  

• Dereplication:

  halo dereplicate -o <project_path> [-g <GNPS_folder>] [-ud <user_database.csv>]
  

• Create training dataset:

  halo create-ds <project_path> [-c <config_file>]
  

• Train model:

  halo train <project_path> [-c <config_file>] [-m search]
  

If you need to modify configuration parameters, edit the config file (download it here) and override the default settings by specifying:

-c [user_config_file]

See documentation for more applications.

License

This code repository is licensed under the MIT License.