Aligning Physiological and Instrumental Sensitivity in AI-Driven Volatilomics via GC×GC–HRMS

The Artificial Intelligence (AI) smelling machine, developed within the sensomics framework, is grounded in the quantitative determination of odorants to reconstruct a chemical odor code via odor activity values (OAVs). Its analytical robustness depends on the alignment between physiological sensitivity—defined by human olfactory perception thresholds—and instrumental sensitivity. The central methodological challenge is therefore to quantify compounds at concentrations approaching or below ng·kg⁻¹ levels while maintaining chromatographic resolution, signal integrity, and identification reliability. Comprehensive two-dimensional gas chromatography (GC×GC) with thermal modulation provides a decisive advantage. Thermal modulation induces enhanced band compression in space, increasing signal-to-noise ratios and peak detectability. The structured separation space of GC×GC markedly reduces co-elution in highly complex volatilomes, expanding peak capacity and preserving trace-level analytes that would be hidden in one-dimensional GC. This gain in separation efficiency directly improves quantitative accuracy at concentrations relevant to olfactory thresholds. Coupling GC×GC with high-resolution mass spectrometry (HRMS) further strengthens analytical performance. Accurate mass measurement and high mass resolving power enhance molecular selectivity, improve deconvolution in dense chromatographic regions, and increase confidence in structural annotation. The orthogonality of two-dimensional separation combined with HRMS discrimination enables reliable quantification of low-abundance odorants critical for OAV calculations. Upstream, high-concentration capacity headspace sampling techniques—such as SPME Arrow and in-tube extraction (ITEX)—increase analyte preconcentration and system loading while preserving chromatographic performance. The integrated strategy of HCC sampling, thermally modulated GC×GC, and HRMS detection effectively lowers the instrumental detection boundary toward physiologically relevant odor thresholds. Application to complex matrices such as peanuts (Arachis hypogea) demonstrates how multidimensional separation and high-resolution detection enable a more faithful reconstruction of the food chemical odor code.