Mining Structural Information from Gas Chromatography-Electron-Impact Ionization-Mass Spectrometry Data for...

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Toxicology

Background:

• Quantitative structure-activity relationship (QSAR) models traditionally rely on molecular descriptors.
• Machine learning advances enable QSAR prediction using analytical signals, such as mass spectra, bypassing the need for complete structural elucidation.
• Interpreting the complex relationship between mass spectra and chemical structure remains a challenge for developing advanced QSAR methods.

Purpose of the Study:

• To determine if gas chromatography-electron-impact ionization-mass spectrometry (GC-EI-MS) data provide sufficient structural information for QSAR prediction.
• To compare the predictive performance of QSAR models based on analytical signals versus traditional molecular descriptors.
• To evaluate the utility of analytical-signal-based QSAR for predicting physicochemical properties and toxicities of unknown compounds.

Main Methods:

• Developed QSAR prediction models using machine learning with full scan mass spectral data as input.
• Compared analytical signal-based QSAR with traditional QSAR using four molecular descriptors: ECFP6, CDK topological descriptor, MACCS key, and PubChem fingerprint.
• Evaluated predictive performance for molecular weight, log K_o-w, boiling point, melting point, water solubility, and oral toxicities in rats and mice.

Main Results:

• Both analytical and molecular descriptors captured structural information, albeit differently, with comparable predictive performance.
• The analytical-descriptor-based QSAR approach successfully predicted physicochemical properties and toxicities for structurally unknown chemicals.
• This analytical-descriptor-based approach extended predictive capabilities beyond the scope of molecular-descriptor-based methods.

Conclusions:

• GC-EI-MS data contain meaningful structural information that can be leveraged for QSAR prediction.
• QSAR models based on analytical signals are valuable for assessing unknown chemicals, offering a complementary approach to traditional methods.
• The analytical-signal-based QSAR approach shows significant promise for broader applications in chemical safety and discovery.