Researchers at the University of Campinas (Campinas, Brazil) and Waters Research Centre (Budapest, Hungary) have introduced a rapid and automated method with fingerprint coffee samples and predictive sensory properties using laser-assisted rapid evaporation ionization mass spectrometry (LA-REIMS) with high-resolution mass spectrometry. Using artificial neural networks (ANN), this method achieved high prediction accuracy (87-96%), superior to traditional PLS-DA models. Major contributors to sensory perception include sugars, chlorogenic acids and fatty acids. This approach reduces dependence on subjective sensory panels and includes new algorithms to interpret the weights of the ANN model to better understand compound associations. LCGC International I spoke to Leandro Wang Hantao of Campinas University about their work and the papers that arose from it (1).
What motivated the development of an automated laser-assisted rapid evaporation ionization mass spectrometry (LA-REIMS) system for coffee sensation assessment?
The Brazilian Coffee Cooperative has sought to evaluate the latest technology to improve its quality control process. Many colleagues worked with them to study a variety of analytical techniques. Our specific focus was identifying alternative methods for estimating coffee quality. The goal was not to replace professional tasters, but to provide reliable tools for everyday quality monitoring.
We found an interesting article on laser-based desorption and ionization techniques, requiring a rapid method to fingerprint analytes related to coffee aroma. This technology appears to be ideal for rapid analysis, and when combined with high resolution mass spectrometry, it provided the potential for information-rich fingerprints. I discussed this instrumental approach with a colleague at Waters Corporation. This prototype utilized laser-assisted rapid evaporation ionization (LA-REIMS) in a 96-well plate format to allow for individual analysis within 10 seconds. This throughput met the requirements of the cooperative.
As far as we know, the main advantage of La-Reims is speed. Chromatography-based workflows struggle to meet this pace. However, each analytical approach has distinct characteristics. In this project, speed was prioritized over sensitivity and resolution.
Can you explain why minimal sample preparation is advantageous for high-throughput coffee analysis?
Sample preparation is essential for analysis methods. However, the degree and complexity depend on the scope and methodology of the analysis employed. For laser-assisted rapid evaporation ionization (LA-REIMS), preparations were included to grind coffee for representative sampling and add moisture to promote laser absorption. This minimal preparation was key to achieving rapid analysis. In contrast, equilibrium-based techniques such as solid phase microextraction (SPME) are required at least several minutes for sample preparation, even under non-equilibrium conditions.
Why is it important to integrate both sensory panel results and instrumental analysis into coffee quality assessments?
Instrumental analysis should be integrated with chemical metrics to support routine quality control for coffee sensation assessment. This combination allows for reproducible and reliable results. Sensory panels train initial predictive models and are essential for continuous validation and improvement, which improves the accuracy and over time of the model. This continuous effort is required to build an extensive, representative database.
This study achieved a predictive accuracy of 87-96% of coffee sensation properties. What do these results mean about the reliability of the Artificial Neural Network (ANN) approach in LA-Reims and the industrial environment?
Given that this is the initial model, the predicted accuracy of the sensory panel achieved results equivalent to about 80%. Continuing improvements are expected to further improve the accuracy of the chemical measurement models.
Why does the Ann model partially outperform the least squares classification analysis (PLS-DA) in predicting coffee sensation properties? Can you explain in detail the advantages and limitations of ANN compared to linear models like PLS-DA?
I particularly value the proverb “performing common and unusual things.” Although linear models have proven successful in most applications, we may observe nonlinear relationships between dependent and independent variables, apparent in residual patterns and systematic prediction errors. In these cases, it is recommended to carefully select alternative methods (artificial neural networks, ANNs, etc.).
How has a new algorithm been developed to evaluate mass-to-charge ratios?m/z) Does the importance of the ANN model increase the interpretability of machine learning and provide results in food analysis?
A key concern with machine learning is the inherent lack of transparency in the way models utilize input data. Factor-based chemosensing addresses this by allowing the evaluation of combinations of independent variables (load and regression vectors). This new algorithm provides an essential tool for assessing the chemical validity of predictive models using ANN.
Compounds such as sugars, chlorogenic acids, and fatty acids have been tentatively identified as affecting coffee sensation attributes, why are they important in coffee chemistry?
Profiling AROMA-related compounds via GC×GC-MS and LC-HRM effectively reveals the link between chemical profiles and sensory attributes, while ANN models work through direct and primarily indirect correlations within the mass spectral fingerprint. As a result, interpreting sensory predictions is less simple than what is desired.
How can the combination of mass spectrometry and machine learning reconstruct the quality control process in the broader food industry?
I think it's worth using instrumental analysis and chemical metrics for daily coffee quality control. Professional tasters can help train and refine the model and free the model to focus on product development and innovative work.
Is the prototype LA-REIMS system commercialized?
Working with the prototype presented an attractive challenge, but we concluded that the technology is mature and market ready. It is currently available commercially through Ambimass KFT.
How does this study think about it in ways that will affect coffee production, consumer satisfaction, and even the future of coffee marketing strategies?
Detailed and actionable data is fundamental to ensuring product consistency and, importantly, fundamental to driving product innovation.
What challenges could arise when deploying this system commercially, particularly with regard to equipment costs, operator training, or data interpretation?
Process and product innovation requires investment in human resources and infrastructure, and equipment costs become a secondary consideration. Operator training is essential to ensuring reliable measurements and maximizing equipment uptime. Data interpretation is streamlined for routine analysis. Unlike complex model development, results are easily visualized via QC charts and graphs, with immediate reporting.
How do you see the interpretability of the ANN model bridging the gap between analytical chemistry and human sensory analysis in the coffee industry through new algorithms?
It provides an essential tool for deciphering nonlinear relationships in chemosensory studies.
Can the methodology presented here be adapted to other foods? If so, what adjustments might be needed?
This workflow is readily transferred to other foods, and only minor adjustments are required, particularly in the sampling and sample preparation protocols, to ensure aliquot representation.
reference
- Kelis Cardoso, VG; Balog, J. ; Zsellér, V. ; Karancsi, T. Sabin, GP; LW prediction of coffee properties by artificial neural networks and laser assisted rapid evaporation ionization mass spectrometry. Food Res. int. 2025, 203, 115773. doi:
10.1016/j.foodres.2025.115773
