ChloroDBPFinder

ChloroDBPFinder utilizes machine learning to recognize chlorinated disinfection byproducts (Cl-DBPs) from nontargeted LC-HRMS analysis of water samples.

Key Features:

• Hierarchical Machine Learning Framework: Two random forest-based models classify Cl-containing chemicals and annotate the number of chlorine atoms present.
• Cl-Containing Chemical Recognition: Binary classifier identifies Cl-containing molecular formulas from LC-HRMS data.
• Chlorine Number Annotation: Multiclass classifier determines the number of Cl atoms in recognized Cl-containing molecular formulas.
• Known and Unknown Compound Annotation: Annotates both known and unknown Cl-containing compounds from LC-HRMS data.
• Molecular Networking Integration: Supports tentative structural annotation of Cl-DBPs via molecular networking.
• Streamlined LC-HRMS Data Processing: Processes liquid chromatography-high-resolution mass spectrometry data for Cl-DBP feature extraction.

Scientific Applications:

• Drinking Water Disinfection Byproduct Analysis: Identifies Cl-DBPs in chlorinated tap water samples, as demonstrated with aspartame chlorination data sets.
• Natural Organic Matter Chlorination Studies: Extracts and annotates Cl-DBPs from chlorinated humic substance samples.
• Environmental Monitoring: Enables nontargeted identification of Cl-containing compounds in complex environmental water matrices with diverse chemical backgrounds and analytical fluctuations.

Methodology:

ChloroDBPFinder implements a hierarchical machine learning framework with two random forest-based models trained on approximately 1.4 million distinctive molecular formulas from PubChem. The first layer is a binary classifier for recognizing Cl-containing chemicals, and the second layer is a multiclass classifier for annotating the number of Cl atoms present. The model was evaluated on over 14,000 unique molecular formulas from NIST20.