Development of Unknown Compounds Analysis Method combining High-Resolution Mass Spectrometry, Soft Ionization Technique and AI Technology for Comprehensive 2-Dimensional Gas Chromatography

Significance of the Topic

Comprehensive identification of unknown organic compounds is critical across environmental monitoring, petrochemical analysis and metabolomics. Traditional GC/MS and GCxGC/MS workflows rely on spectral libraries like NIST, which cover only a fraction of known chemicals. Integrating high‐resolution mass spectrometry, soft‐ionization techniques and AI–driven structure elucidation addresses this gap, enabling rapid, automated characterization of novel analytes.

Aims and Overview of the Study

This work presents a new qualitative analysis strategy for unknown compounds by combining:

• Integrated molecular‐formula determination using high‐resolution GC-TOFMS with EI and soft‐ionization modes.
• AI‐based structural elucidation leveraging a 130 million‐entry predicted EI mass spectral database (“AI Library”).

Applications to GCxGC workflows demonstrate efficient data handling, library searching and substructure annotation.

Methodology and Workflow

1. Sample Introduction and GCxGC Separation
   Two‐dimensional GC with a thermal modulator achieves enhanced separation of complex mixtures; retention indices calibrated with n-alkanes.
2. High‐Resolution MS and Soft-Ionization
   JEOL JMS-T2000GC combines EI with Field Ionization (FI) or Photo-Ionization (PI) to capture both fragment and molecular ions with <1 ppm mass accuracy and resolving power ~30 000 @ m/z 614.
3. Integrated Qualitative Analysis
   Chromatographic peak detection co-registers EI and soft-ionization data. Library‐based matching (match factor, retention index) narrows candidate molecular formulas; fragment‐ion analysis further refines formula coverage.
4. AI Structure Analysis
   Deep‐learning models predict EI spectra and retention indices for 130 million PubChem structures. Candidate ranking by cosine similarity and AI score yields top structural formulas and substructure predictions.

Used Instrumentation

• GC-TOFMS: JMS-T2000GC (JEOL)
• Thermal Modulator: INSIGHT (SepSolve), cryo-free cooling
• GC Columns: BPX5 (1st), Rxi-17Sil MS (2nd)
• Ion Sources: EI/FI comb. source, EI/PI comb. source

Main Results and Discussion

• Construction of a 130 million predicted EI spectrum database with associated RI predictions.
• AI model performance: average cosine similarity 0.80; 87 % of test compounds yielded correct or similar top‐rank structures.
• Case study (ID #187): unknown compound identified via integrated workflow; AI structural analysis provided high‐scoring candidates consistent with measured spectra.
• GCxGC data handling: automated peak detection, library search and categorization of >700 compounds in sake metabolomics.

Benefits and Practical Applications

The proposed solution enables automated, high‐throughput identification of non‐library compounds in environmental, petrochemical and food analysis. Combining accurate formula assignment with AI spectral matching reduces expert intervention and accelerates discovery of novel analytes.

Future Trends and Possibilities

Continued expansion of AI‐predicted spectral libraries and substructure models will improve coverage of specialized chemical classes (e.g. polymers, biogenic metabolites). Integration with real‐time data acquisition and cloud‐based platforms may deliver on‐the‐fly unknown compound annotation for field applications.

Conclusion

By uniting high‐resolution GC-TOFMS, soft‐ionization, and AI structural analysis, this workflow offers a robust solution for comprehensive unknown compound identification across diverse GC/MS fields.

Reference

• M. Ubukata et al., Rapid Commun. Mass Spectrom. 34 (2020). DOI: 10.1002/rcm.8820
• A. Kubo et al., Mass Spectrometry 12 (2023) A0120.