Construction of a Comprehensive Contribution Ranking Model for Baijiu Aroma Compounds Based on Multi-Source Molecular Feature Fusion and Validation by Aroma Recombination.

Baijiu aroma arises from the combined effects of diverse volatile and semivolatile trace components within the Baijiu matrix. A single concentration value or odor activity value (OAV) cannot fully characterize the relative importance of a compound in a specific sample. A knowledge base containing 5572 Baijiu aroma-related compounds was used to develop a comprehensive contribution ranking model and a sample-level intensity prediction model. Odor evidence, molecular structure, physicochemical properties, volatility and partitioning characteristics, matrix information, and evidence quality were integrated. Comprehensive contribution is defined here as a within-sample relative ranking index obtained from multi-source molecular features, concentration information, and matrix conditions under a specified model setting. It is not a direct measurement of the true sensory contribution of a compound. After data standardization and screening for feature computability, 5568 compounds were included in the scorer. The R2, RMSE, and MAE obtained from 9000 records in the formal independent test set were 0.952018, 0.045684, and 0.036464, respectively. Light-aroma, strong-aroma, and sauce-aroma Baijiu were then analyzed by GC×GC-TOFMS. A total of 732 trace components were obtained, of which 191, 474, and 510 were detected in the three sample groups, respectively. The aroma-related compounds in each sample group were ranked within the group. The top 20 compounds were selected and recombined at their semiquantitative concentrations in the original Baijiu. Sensory evaluation by 10 assessors showed overall similarity scores of 8.70 ± 1.16 and 9.20 ± 1.03 for the light-aroma and strong-aroma recombination samples, respectively. The correlation coefficients between their 10-dimensional sensory profiles and the mean profiles of the original samples were 0.988 and 0.992. The overall similarity score of the sauce-aroma recombination sample was 5.40 ± 1.26. Significant proportional deviations were observed among acidic aroma, sauce-like and roasted aroma, and floral and fruity aroma (p < 0.05). These results indicate that multi-source feature fusion can provide an interpretable basis for the relative screening and experimental prioritization of candidate aroma compounds in Baijiu. The rankings for the light-aroma and strong-aroma samples received strong sensory support. However, they should not be interpreted as absolute contributions across aroma types. Evaluation of sauce-aroma Baijiu requires further incorporation of aroma-type-specific prior information, low-OAV components, component interactions, and effects of the authentic Baijiu matrix.