Machine-learning-assisted screening of key flavor compounds in pumpkins.
Jiangfei Chai1,2, Xiangdong Zhang3, Jincha Cao3
1College of Food and Health, Zhejiang Agriculture and Forestry University, Hangzhou, China.
Journal of the Science of Food and Agriculture
|June 4, 2026
Summary
Machine learning identified key volatile organic compounds (VOCs) in raw and steamed pumpkins. These aroma markers, like alcohols and 2,3-butanedione, explain distinct pumpkin flavors, aiding breeding and processing.
Area of Science:
- Food Science
- Analytical Chemistry
- Sensory Science
Background:
- Pumpkin flavor is crucial for consumer acceptance and market value.
- Volatile organic compounds (VOCs) significantly influence food aroma.
- Machine learning (ML) can model complex relationships between VOCs and sensory perceptions.
Purpose of the Study:
- To identify the distinct flavor profiles of raw and steamed pumpkins.
- To pinpoint specific VOC markers responsible for these sensory attributes.
- To apply ML for linking chemical composition to sensory data.
Main Methods:
- Sensory evaluation of raw and steamed pumpkin samples.
- Gas chromatography-time-of-flight mass spectrometry (GC-ToF/MS) for VOC profiling.
- Integration of ML algorithms (e.g., ExtraTrees) and SHAP analysis to identify key VOCs.
Main Results:
- Raw pumpkin aroma dominated by green and cucumber notes (alcohols).
- Steamed pumpkin aroma characterized by chestnut, creamy, sweet, and potato notes (2,3-butanedione, methional).
- ML identified 25 important VOCs, with ExtraTrees showing high performance in linking VOCs to sensory traits.
Conclusions:
- Established a link between pumpkin's sensory profile and its VOC composition.
- Identified key aroma compounds for pumpkin sensomics research.
- Provided a foundation for aroma-focused pumpkin breeding and product development.
