ANALYSIS OF AROMA COMPOUNDS IN SPICES BY COMPREHENSIVE TWO-DIMENSIONAL GAS CHROMATOGRAPHY/TIME-OF-FLIGHT MASS SPECTROMETRY WITH MACHINE LEARNING-BASED STRUCTURE ELUCIDATION AND MOLECULAR FORMULA ESTIMATION

Importance of the Topic

This study addresses critical challenges in aroma compound analysis by comprehensive two-dimensional gas chromatography coupled with time-of-flight mass spectrometry (GCxGC-TOFMS). Accurate identification of volatile constituents in spices is essential for quality control, flavor optimization and authenticity assessment. Traditional workflows relying on NIST database matching and manual formula interpretation often leave many compounds unannotated or ambiguously characterized. Incorporating machine learning for molecular formula recommendation and structural elucidation enhances confidence in compound identification and accelerates analytical throughput.

Aims and Overview of the Study

The primary goal was to apply two novel machine learning–based strategies to the analysis of spice aroma compounds. Method ML1 predicts the elemental composition from soft-ionization mass data, while ML2 infers molecular structure by simulating electron ionization spectra. These approaches were integrated into a GCxGC-TOFMS workflow with solid-phase microextraction sampling to evaluate their performance in real-world spice matrices.

Methodology and Instrumentation

• Sample: 0.25 mg of dried cardamom seeds extracted by SPME at 50°C for 30 min using a DVB/CAR/PDMS fiber.
• Chromatography: Two-dimensional separation with a BPX-5 (30 m × 0.25 mm i.d., 0.25 μm) primary column and an Rxi-17Sil MS (3.4 m × 0.15 mm, 0.15 μm) secondary column. Thermal modulation every 6 s.
• Mass Spectrometry: JMS-T2000GC high-resolution TOFMS (JEOL) operated in electron ionization (EI) and field ionization (FI) modes to acquire accurate mass data.
• Data Processing: msFineAnalysis AI software for peak detection, alignment, NIST DB searching (347,100 entries) and ML-based formula/structure prediction.

Main Results and Discussion

A total of 518 volatile compounds were detected in the cardamom extract. Annotation combined NIST database hits, retention indices, isotope patterns from FI and accurate EI fragments. A case study of an unregistered compound (“Compound A”) demonstrated the workflow: traditional NIST matching suggested Santolina triene (C10H16), but FI data gave a base peak corresponding to the molecular ion and ML1 estimated the formula C10H16O with high confidence. Subsequently, ML2 proposed candidate structures for monoterpene alcohols, outperforming manual spectral interpretation and narrowing down plausible isomers.

Benefits and Practical Applications of the Method

• Accelerated annotation of unknown volatiles not present in standard libraries.
• Reduced ambiguity in molecular formula assignment through ML-driven elemental probability distributions.
• Efficient structural hypothesis generation using AI-predicted EI spectra.
• Enhanced reliability of GCxGC-TOFMS for flavor and fragrance research, QA/QC and authenticity testing.

Future Trends and Applications

Further integration of deep learning models with expanding spectral and structural databases will improve prediction accuracy and broaden compound coverage. Real-time ML-assisted identification in on-line GC-MS platforms and incorporation into lab informatics systems can streamline workflows. Extension to non-volatile or thermally labile analytes by coupling with alternative ionization techniques is also foreseeable.

Conclusion

This study demonstrates the viability of machine learning–based molecular formula recommendation and structural elucidation in GCxGC-TOFMS analysis of spice aromas. The two ML models significantly reduce manual interpretation effort and enable rapid, reliable annotation of unknown compounds, representing a valuable advancement for analytical chemistry and flavor science.

References

1. Kubota A., Kubo A., Ubukata M. Analysis of aroma compounds in spices by comprehensive two-dimensional gas chromatography/time-of-flight mass spectrometry with machine learning-based structure elucidation and molecular formula estimation. JEOL Ltd., Tokyo, Japan.