The first stage, referred to as Food Flavour Analysis 1.0, relied solely on sensory analysis performed by trained experts using human sensory organs to evaluate food characteristics. While sensory analysis provided valuable insights, it was limited by its inability to detect harmful gases in food or to elucidate the mechanisms underlying flavour formation. The lack of instruments during this stage restricted the effectiveness and scope of sensory evaluations 10.

The introduction of instrumental analysis in the second stage, Food Flavour Analysis 2.0, was facilitated by technological advancements during the second industrial revolution. Techniques such as chromatography and sensor technologies allowed researchers to study volatile and soluble components of food and beverages in greater depth. Instruments such as GC and LC, combined with detectors like mass spectrometry (MS), ion mobility spectroscopy (IMS), flame ionisation detectors (FIDs), and thermal conductivity detectors (TCDs), enabled precise flavour profiling. Additionally, E-nose and E-tongue provided specific flavour fingerprints with high reliability, complementing or even replacing GC in some cases. However, these technologies were sensitive to environmental changes, and their responses to compounds with specific functional groups were sometimes weakened, limiting their effectiveness during manual identification 10.

The third stage, Food Flavour Analysis 3.0, marked the integration of sensory and instrumental analyses, leading to the development of molecular sensory science. Techniques such as GC-MS and GC-O were combined with odour activity value (OAV) measurements, omission tests, and aroma reconstitution experiments to quantify, describe, and elucidate the molecular basis of key flavours at a deeper level. Despite these advancements, the results of olfactometry analyses were often dependent on sensory evaluators, making it difficult to establish uniform standards 10.

The fourth stage, Food Flavour Analysis 4.0, is characterised by the application of machine learning in the food industry. Machine learning algorithms can identify patterns, recognise similarities, and differentiate variations in sample characteristics. Supervised learning algorithms, when combined with traditional flavour analysis techniques, enable the prediction of flavours in unknown food samples with greater speed and objectivity than manual methods. Furthermore, machine learning facilitates the processing of large datasets, reducing analysis time and improving efficiency. It can identify complex patterns in high-dimensional variable spaces and automatically optimise algorithms to predict outcomes based on new data 11. However, there remains a need for further research to summarise the advances in using different machine learning algorithms as classification criteria for food flavour analysis 10.

In addition to machine learning, Industry 4.0 technologies such as real-time data transmission and cloud-based platforms further enhance flavour analysis capabilities. These systems enable continuous monitoring and seamless sharing of analytical data across laboratories, facilitating instrument synchronisation and large-scale data integration. Such connectivity allows analytical outputs from techniques like GC–MS and electronic noses to be integrated into centralised databases for rapid interpretation and model training. Consequently, the combination of advanced analytics and interconnected systems supports the development of more efficient, reproducible, and predictive flavour analysis workflows 10 11.

The primary sensory analytical technique is descriptive sensory analysis, which is performed by trained sensory panels. Descriptive analysis applies a flavour lexicon, a set of words used to describe the flavour of a product based on the sensory physiological perception of the panels. The qualitative descriptions include the appearance, odour, taste, and texture of the product 5. Two other sensory analytical techniques include discrimination tests, which assess whether products are similar or different in terms of general or specific characteristics, and consumer tests, which focus on the hedonic and emotional responses of consumers towards products to understand consumer acceptability 12.

Taste perception can differ between individuals, including trained sensory panel members. Both flavour and texture perception influence how taste is experienced. Flavour is affected by a combination of taste, aroma, and chemesthetic sensations, which can also be influenced by cognitive descriptions and attention 5. While taste primarily encompasses sweet, sour, bitter, salty, and umami, aroma adds layers of complexity, distinguishing between floral, fruity, earthy, spicy, and countless other descriptors. Texture plays a crucial role in defining sensations such as creaminess, crunchiness, or chewiness, adding to the mouthfeel of a product. Moreover, individual preferences, genetic variations in taste perception, and cultural upbringing further shape how flavours are perceived and interpreted, often introducing a degree of cognitive bias.

Flavour descriptions frequently rely on metaphors, serving as conceptual guides for both spoken and written accounts. Within the domain of flavour analysis, various metaphorical frameworks are employed, including discourse metaphors that perceive flavour as an event, a dimensional entity, a balance, a living entity, or an art form 13. For example, wine connoisseurs may use terms such as "oaky", "buttery", or "fruity" to articulate the characteristics of a particular vintage. Similarly, chefs might describe a dish as "savoury", "spicy", or "succulent", capturing its flavour profile and mouthfeel. In addition to standard descriptors, metaphors, analogies, and cultural references are frequently used to evoke specific sensations and emotions associated with flavours. Categorising flavours is a complex task that involves classification based on various criteria, such as taste, aroma, origin, and culinary use. The flavour wheel, a visual representation of flavour categories, is a common tool used by professionals in the food and beverage industry to organise and comprehend the vast spectrum of flavours 14. Furthermore, describing complex flavours often requires a rich and diverse vocabulary encompassing a wide range of taste, aroma, and texture descriptors. The effectiveness of teaching methods may vary in influencing one's ability to analyse and describe complex flavours.

Sensory techniques can be coupled with instrumental flavour analysis methods, such as GCO. This combination can elucidate the relative impacts of many volatile compounds on flavour characteristics 15. The E-nose and E-tongue are instrumental analytical techniques designed to mimic human sensory perception, specifically olfaction and taste. The E-nose is used to analyse volatile food flavour compounds and reveal aroma differences in food while eliminating operator fatigue, as the instrument can obtain repeatable measurements. The E-nose comprises a group of chemical sensors associated with a pattern-recognition system that detects and processes odours 16 15 7. The sensors exhibit varying sensitivity to different odours due to chemical interactions between compounds and sensors. Changes in physical or chemical properties result in electrical signals recorded by a computer system. These signals are then analysed using chemometric tools, and when the sensor compares patterns from a series of samples, the acquired differences can be correlated with perceived aromas 15.

The E-tongue is based on the tongue’s ability to differentiate between distinct taste qualities, such as sourness, saltiness, sweetness, bitterness, and umami. E-tongue is non-specific sensors composed of sensing elements, including enzymes, lipids, and metallic particles, which have low selectivity but can analyse and differentiate between liquid samples based on their unique characteristics 17 16. The principle behind the E-tongue involves combining signals from specific, non-specific, and overlapping sensors with pattern recognition when detecting polyphenols and predicting sensory attributes. Similar to human taste receptor cells, amperometry sensors composed of metals, conducting polymers, phthalocyanine films, and biosensors analyse the sensory properties of food instead of sensory receptors. The sensors then release data on food properties, which is ultimately interpreted as perceived taste 18 .

GC is a separation technique used to analyse volatile and semi-volatile compounds. In GC, analytes are vaporised and transported by an inert carrier gas through a capillary column containing a stationary phase, where compounds are separated based on their volatility and interactions with the column phase. The components of a sample are separated by dissolving and vaporising them within the GC system, using two phases: the stationary and mobile phases 19. The advantages and disadvantages of GC are summarised in Table 1.

Table 1. Advantages and disadvantages of gas chromatography

MS is a powerful technique used for identifying novel molecules, quantifying known compounds, and elucidating the structural and chemical properties of various substances 25. Mass spectrometry (MS) is an analytical technique that identifies compounds by measuring the mass-to-charge ratio of ionised molecules. In MS analysis, compounds are ionised, fragmented, and separated in a mass analyser, generating characteristic mass spectra that enable compound identification. The advantages and disadvantages of MS are summarised in Table 2.

Table 2.Advantages and disadvantages of mass spectrometry

GC-MS plays a crucial role in flavour analysis, offering unmatched capabilities for identifying and characterising volatile and semi-volatile flavour compounds. GC separates complex mixtures by exploiting differences in boiling points, allowing individual flavours to be analysed sequentially 31. Once separated, the compounds enter the MS system, where they are ionised and fragmented to generate unique mass-to-charge ratio fingerprints for identification 32, as shown in Figure 2. This synergy enables precise identification of specific flavour compounds and their relative abundance.

In flavour analysis, volatile compounds are often extracted prior to GC-MS analysis using techniques such as SPME or SBSE. These extraction approaches allow the analysis of both known aroma compounds and broader volatile profiles present in food samples. However, the complexity of food matrices can present analytical challenges, including matrix effects, co-elution of compounds, and uncertainties in compound identification. As a result, confirmation of compound identities often requires comparison with reference standards or retention indices 7.

Figure 2. The GC-MS instruments

GC–MS and liquid chromatography-based techniques differ in their analytical principles and typical applications. Both techniques rely on pressure-driven flow for separation; however, GC uses an inert carrier gas to transport vaporised compounds through the column, whereas HPLC employs high-pressure liquid solvents to achieve separation. GC–MS integrates chromatographic separation with mass spectrometric detection for volatile compounds. Similarly, liquid chromatography can also be coupled with mass spectrometry (LC–MS) for compound identification, particularly for non-volatile and thermally labile compounds. In contrast, HPLC is primarily a separation technique and is commonly coupled with detectors such as ultraviolet (HPLC–UV) or mass spectrometry (LC–MS) for compound identification and quantification. These LC-based systems are particularly suitable for analysing non-volatile, thermally labile, or highly polar compounds in complex food matrices 31.

Since chromatography itself performs compound separation, reliable compound identification typically requires coupling with appropriate detection systems such as UV detectors or mass spectrometry. Due to these differences, GC-MS is widely used for analysing volatile aroma compounds like aldehydes, esters, and terpenes found in essential oils, fruits, and spices. Meanwhile, HPLC-based systems are better suited for analysing non-volatile or thermally labile taste compounds dissolved in liquid solvents prior to analysis, including organic acids, polyphenols, and amino acids, which contribute to sensory attributes such as bitterness, astringency, sweetness, and umami 37.

In flavour analysis, GC-MS is ideal for detecting volatile compounds in fresh and processed foods, while HPLC is preferred for non-volatile taste components. For complex food matrices, both techniques are often combined to achieve a complete flavour profile 38. Given the diversity of available platforms, a comparative evaluation of their performance metrics is essential for informed method selection.

In practical flavour studies, volatile compounds are commonly extracted prior to GC-MS analysis using techniques such as SPME or SBSE. These approaches allow efficient concentration of volatile molecules from complex food matrices before chromatographic analysis. For example, in coffee aroma research, SPME coupled with GC-MS has been widely applied to characterise volatile compounds responsible for coffee aroma. Studies have identified furans and pyrazines as the predominant classes of volatile compounds contributing to the characteristic roasted, caramel, and nutty aroma notes of coffee 39.

For both research and industrial flavour analysis, a careful comparison of instrumental platforms is necessary to assist well-informed method selection. Their usefulness is significantly impacted by differences in sensitivity, reproducibility, cost, analytical throughput, and suitability for volatile and non-volatile substances. A comparison of critical performance metrics for the main analytical platforms used in flavour profiling is shown in Table 4.

Table 4.Comparative evaluation of instrumental platforms for flavour analysis

GC-MS remains the benchmark technique for volatile aroma analysis due to its high sensitivity, comprehensive spectral libraries, and strong separation capability for complex mixtures 22. However, column ageing, detector drift, and variations in sample preparation techniques, especially during solid phase microextraction fibre conditioning, may have an impact on reproducibility. Although the identification of trace-level aroma-active chemicals frequently necessitates derivatisation, which increases analytical complexity and processing time 43, HPLC remains widely used for the analysis of non-volatile taste-related compounds, including polyphenols, amino acids, and organic acids 37. Ambient ionisation mass spectrometry techniques such as DART and DESI enable rapid chemical profiling with minimal sample preparation. These techniques ionise compounds directly from sample surfaces under atmospheric conditions, allowing fast screening of food samples. However, their applications in flavour analysis are often limited by lower quantitative reproducibility and dependence on spectral databases for compound identification 44.

E-nose and E-tongue are frequently used for real-time monitoring in industrial settings because they provide quick, high-throughput screening 15. Although they have potential for process control due to their ease of deployment and comparatively low cost, sensor responses are sensitive to environmental conditions such as temperature and humidity. Furthermore, the need for regular recalibration limits cross-laboratory standardisation and lowers long-term stability 45.

Overall, this comparative evaluation demonstrates that no single analytical platform is universally optimal for flavour profiling. More accurate and thorough flavour characterisation is frequently produced by integrated analytical techniques, such as combining targeted GC-MS confirmation with electronic sensor screening. In industrial applications, practical considerations such as operational cost, staff training requirements, and analysis time frequently outweigh maximal analytical resolution. These factors emphasise the significance of structured method-selection frameworks that strike a compromise between analytical rigour and practical feasibility when dealing with complex food matrices.

While each analytical technique offers distinct advantages, their suitability is highly dependent on the intended application. For rapid screening and real-time monitoring, E-nose systems provide fast and cost-effective analysis, although they lack compound-specific identification. In contrast, GC–MS remains the preferred approach for regulatory and confirmatory analysis due to its high sensitivity and ability to identify individual volatile compounds. HPLC-based techniques are more suitable for non-volatile and thermally labile compounds. Therefore, the optimal analytical strategy is not universal but application-driven, requiring a balance between speed, sensitivity, cost, and analytical specificity 22.

Building on these instrumental foundations, the integration of chemometrics and machine learning represents a shift from descriptive to predictive flavour analysis 46. Principal Component Analysis and Partial Least Squares Regression are two chemometric techniques that make it possible to uncover significant patterns from complex datasets generated by chromatography-based methodologies and electronic sensing platforms 47. These methods improve classification, prediction, and feature selection when paired with machine learning algorithms such as Random Forest, Support Vector Machines, and neural networks.

Key aroma active chemical identification, sensory score prediction, and geographical origin authentication are examples of practical applications. For example, by identifying volatile indicators, machine learning models applied to GC-MS data have successfully categorised olive oil sources with above 95% accuracy 48. Unsupervised methods help to visualise sample groupings and spot anomalies, whereas supervised approaches correlate chemical profiles with sensory qualities. Predictive flavour analysis typically follows a structured workflow involving the acquisition of analytical data, such as GC–MS spectra or sensor responses, followed by data pre-processing and feature extraction prior to model development. Machine learning algorithms are then trained and validated using independent datasets before being applied to predict the flavour profiles of new samples. However, accurate prediction remains challenging due to the inherently multimodal nature of flavour, where chemical composition, sensory perception, and human subjectivity are closely interconnected. Unlike other omics fields, flavour perception is influenced by complex interactions among compounds, making prediction less straightforward and requiring the integration of both chemical and sensory datasets to improve reliability 46.

Despite these advancements, difficulties remain. machine learning models frequently demand big, curated datasets, which are limited in flavour science. Industrial adoption is hampered by problems with cross-platform reproducibility, overfitting with limited samples, and model interpretability (commonly referred to as "black box" effects) 49. Although machine learning has been increasingly applied in flavour analysis, the issue of model interpretability remains a significant challenge. To address the “black box” nature of complex models, interpretability approaches such as SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) have been introduced, allowing the contribution of individual variables to model predictions to be examined more clearly 50. Moreover, the selection of machine learning algorithms should be aligned with the type of flavour data being analysed. Deep learning approaches are generally more suitable for handling high-dimensional datasets such as GC–MS spectra, whereas conventional models, including support vector machines and random forests, are often more effective for structured datasets such as electronic nose sensor responses 50. Therefore, appropriate model selection should consider data complexity, interpretability requirements, and the specific objectives of the analysis. Improved data sharing, explainable AI frameworks, and hybrid modelling techniques that incorporate sensory, chemical, and processing elements are essential for future advancements. Overall, these analytical and computational approaches collectively demonstrate the transition towards a more integrated and data-driven framework in flavour analysis.

To summarise, flavour analysis plays a critical role in ensuring the quality, safety, and sensory attributes of food products. By employing a variety of analytical techniques, such as gas chromatography-mass spectrometry, scientists can identify and quantify volatile compounds responsible for flavour and aroma, ultimately enhancing flavour profiles and increasing consumer satisfaction. However, the reproducibility of these analyses across multiple laboratories and experimental setups is still a considerable difficulty 54. Comparative research and industrial quality control are hindered by inconsistent results that are frequently caused by variations in sample preparation, equipment calibration, data processing, and operator experience.

Differences in SPME fibre conditioning, GC column ageing, detector sensitivity drift, and the absence of widely recognised reference materials for complex food matrices are major causes of irreproducibility. Even well-known techniques like GC-MS and HPLC produce data that is challenging to compare or validate among research groups in the absence of standardised methodologies 55. This is especially difficult when creating large-scale flavour databases for machine learning applications and in regulatory contexts. The development of standardised benchmark datasets is hindered by several technical challenges, including matrix effects, variability in sample preparation, and the instability of reference standards. Differences in instrumentation and analytical conditions across laboratories further contribute to reproducibility issues, limiting the comparability of results 56.

To solve these challenges, numerous practical solutions are proposed. First, cross-laboratory calibration and technique validation would be made possible by globally recognised benchmark datasets that include well-characterised food extracts with validated concentrations of important flavour constituents 57. Second, academic, business, and regulatory parties should work together to create standard operating procedures (SOPs) for data reporting, instrument settings, and sample preparation. Third, open-access, quality-controlled flavour databases supplemented with metadata on analytical circumstances would increase transparency and data reuse.

Furthermore, the integration of artificial intelligence technologies for automated quality control, such as signal alignment, predictive calibration, and outlier identification, may reduce human-related variability and enhance flavour profiling's overall dependability 58. Therefore, emphasising on standardisation and reproducibility will aid in the creation of frameworks for flavour analysis that are more reliable, compatible, and applicable to industry.

In order to lessen ecological impact, flavour analysis is increasingly using green analytical chemistry principles as part of the growing emphasis on environmental responsibility. Sustainable techniques aim to preserve analytical performance while reducing the usage of solvents, energy, and waste.

Solvent-free microextraction techniques, such as SPME and SBSE, are important green technologies that eliminate or significantly minimise the need for organic solvents 59. Compared to traditional petroleum-based solvents, bio-based solvents made from renewable resources are more environmentally friendly. Furthermore, ambient mass spectrometry methods save energy and solvent use by enabling quick analysis with minimal sample preparation 41.

By identifying the flavour characteristics in waste streams and byproducts, flavour analysis also helps to value upcycled food ingredients and promotes circular food economies. To ensure that flavour profiling develops in an environmentally friendly way, future advancements should keep including sustainability indicators into technique validation 60.

Although green analytical techniques such as solid-phase microextraction (SPME) reduce solvent use and environmental impact, they may present limitations in sensitivity and reproducibility compared to conventional solvent-based extraction methods 61. Additionally, the adoption of sustainable analytical approaches in industry is often constrained by equipment cost, scalability, and compatibility with existing workflows. Therefore, the implementation of green techniques requires careful consideration of the trade-offs between environmental benefits and analytical performance 62.

Flavour analysis is essential for industrial product development, quality control, and innovation, especially in response to growing consumer demand for novel and sustainable foods.

In plant-based meat development, flavour optimisation focuses on key volatile compounds such as aldehydes (e.g., hexanal), ketones, and pyrazines, which contribute to desirable meaty and roasted notes. GC–MS has been widely used to identify undesirable compounds associated with lipid oxidation and plant-derived off-flavours, such as beany or grassy notes. These analytical findings enable targeted reformulation strategies, including ingredient modification and processing optimisation, to suppress off-flavours and enhance desirable aroma profiles. Such improvements have been reported to enhance sensory acceptance and consumer preference of plant-based products 63. Similarly, in functional food systems, plant-derived bioactive compounds such as phenolics and essential oils are utilised not only for their antioxidant activity but also for their contribution to flavour enhancement and product stability. Their incorporation has supported the development of clean-label products with improved shelf life and consumer appeal 64. These examples demonstrate how analytical techniques directly guide formulation decisions and contribute to improved product quality and market acceptance.

In industrial practice, flavour profiling guides the development of products such as plant-based meats, functional beverages, and clean label foods44. HPLC helps maximise bitterness in functional teas 65, and GC MS is utilised to discover important aroma compounds that mimic meaty notes in plant-based substitutes 66. Real-time quality monitoring in manufacturing is made possible by electronic noses and tongues, which guarantee consistency and early detection of off flavours 67, alongside emerging nanomaterial-based immunosensors for contaminant detection 68.

At both the molecular and processing levels, flavour analysis reveals the chemical and physical foundations of sensory attributes. Studies on coffee roasting track Maillard reaction products like pyrazines and furans, which contribute to roasted flavours 38. In fermented dairy, dynamic changes in esters and acids are monitored to control flavour development. Understanding these pathways allows for precise manipulation during processing, improving both flavour and nutritional quality 69.  Similarly, processing techniques such as enzyme pretreatment and drying can influence flavour retention and release in cereal-based products, affecting overall sensory perception 70.

The integration of databases such as FlavorDB and machine learning models supports these applications by providing predictive tools for flavour pairing and compound discovery 8. Together, these approaches bridge molecular insight with industrial application, enabling the creation of innovative, sensorially appealing, and sustainable food products. Collectively, these developments emphasise the integration of analytical techniques, molecular insights, and computational tools in shaping a more comprehensive and data-driven flavour analysis framework.

From an industrial standpoint, the implementation of advanced analytical techniques requires careful cost–benefit evaluation. Although GC–MS offers high analytical precision, its substantial cost and operational complexity may restrict its use, particularly among small and medium enterprises (SMEs). In comparison, electronic nose systems provide a faster and more cost-effective alternative, albeit with reduced specificity. Therefore, selecting an appropriate analytical approach involves balancing analytical accuracy, cost considerations, and application needs.