An integrated data-driven framework for mapping grapevine berry valorization

Abstract

Global production of grapes and its associated waste streams continue to rise, but providing robust decision-making support for optimum valorization pathways remains a significant challenge. The emission intensity, free fraction, chemical diversity, and market performance indices have been rarely assessed in combination to establish a data-driven framework for grapevine berry valorization. To bridge this gap, a secondary open-access dataset was assessed in this study. This dataset synthesized published literature on volatile organic compound profiles across diverse grapevine berry samples.

• •Pearson correlation analysis conducted between emission intensity and diversity index.
• •t-test and ANOVA analysis performed to assess differences in diversity index across berry types and tissue sources.
• •Market performance analysis conducted to identify high-priority compounds for targeted extraction.
• •Emission intensity, free fraction, diversity index, and market performance indices were integrated into a unified valorization mapping system.
• •A multi-criteria evaluation framework developed for resource recovery decision-making.

Graphical abstract

Specifications table

 

Subject area	Environmental Science
More specific subject area	Biomass waste valorization
Name of your method	An integrated data-driven evaluation framework for mapping biomass waste valorization
Name and reference of original method	None
Resource availability	This study used secondary open-access dataset. The article describing the dataset has been published by Carriero et al [1]. The corresponding data were obtained from open-access dataset available at https://doi.org/10.5281/zenodo.16410077 [2]. The detailed results of the dataset analysis are provided in Supplementary Information (SI).

Background

The transition toward a circular bioeconomy underlines the strategic valorization of agricultural residues in to high value-added chemicals, while simultaneously reducing food waste across the supply chain [3]. Among global agricultural commodities, grapevine is a fruit of great economic importance, with global production expanding 40% since 1962 to reach 77.8 million tons in 2018 [4,5]. This substantial production generates large volumes of waste, specifically grapevine berry waste, which accounts for over 20% of the initial harvested mass [6,7]. Proper management of this waste stream, guided by principals of circular bioeconomy, is crucial to unlocking its potential. This biomass is a rich source of bioactive compounds, particularly volatile organic compounds (VOCs), functioning as key inputs for various industries, e.g. food, pharmaceutical, and cosmetic [8,9]. Realising this underexploited resource can significantly enhance food security and promote sustainable agriculture, serving as the central motivation for research into integrated grapevine berry valorization.

Despite their potential, implementing waste valorization techniques remains challenging due to the complex multicomponent nature of agro-industrial streams [10]. Current valorization practices lack coordination, resulting in isolated decisions that fail to integrate chemical functionality, market potential, production feasibility, and environmental footprint [11]. This critical gap impedes efficient valorization of waste stream, hindering the development of strategic pathways toward high-value applications.

Data-driven decision frameworks are systematic methods that process data using decision-oriented modeling to shift resource management from a descriptive assessment toward actionable execution. Integrated meta-analytical and spatial modelling approach developed by Wang et. al. is a good case in point, which evaluated greenhouse gas emission intensity alongside productivity metrices in global rice cultivation to identify yield-boosting, low-emission practices [12]. Despite these advances, limited attention has been given to detailed grapevine VOC characterization and actionable decision-making to optimize biomass valorization practices.

To address this gap, this study advances a novel perspective on VOC profiling, by establishing it as a decision-support foundation for actionable valorization. By integrating VOCs emission intensity, compound free fraction, chemical diversity, and market performance indices, this work construct a four-route valorization framework. This approach moves beyond descriptive metabolomic profiling to establish a structured data-driven pathway for high-value resource recovery. However, several limitations should be noted. A key limitation of this study is its reliance on a single secondary open-access dataset compiled from literature spanning 1995 to 2022. Consequently, the dataset may not fully capture the natural variability in VOC composition among different grape cultivars. Second, the prioritization framework carries inherent uncertainty, as semi-quantitative market desirability scores were based on relative economic contribution, cross-industry applications, and purification complexities. A further limitation is that the proposed recovery strategies were evaluated primarily on feasibility rather than through formal techno-economic analysis. By integrating compound-level emission, market, diversity, and physicochemical accessibility data, the framework provides a data-efficient means of pre-screening tool that identifies high-priority valorization targets.

Method details

Overall methodological workflow

The proposed framework was developed as a sequential, data-driven procedure to evaluate and map the valorization potential of grapevine berries. The complete procedure consisted of seven consecutive steps, from dataset preprocessing to the generation of the final valorization map. First, the raw dataset underwent preprocessing to examine patterns in VOC emissions across different grapevine cultivars, berry types (red, white), berry tissues (skin, pulp, must), chemical forms (free and bound), and developmental stages according to the BBCH scale. Second, a set of indices, including emission intensity (E), free fraction (F), diversity index (H), and market performance index (MPI), was defined to quantitively characterize the occurrence, abundance, chemical form, diversity, and market potential of VOC subclasses associated with different berry samples. Third, E and F were calculated for the identified VOC subclasses to characterize their relative occurrence and distribution. Fourth, H was calculated to quantify the diversity of VOC profiles across 56 samples, berry parts and types, and to identify samples with greater potential for the generation of diverse valorization pathways. Fifth, MPI was calculated by integrating the relevant market-related attributes of the identified VOCs, thereby providing an estimate of their potential commercial value. Sixth, the resulting indices were integrated to establish an overall valorization potential for each VOC subclass. Finally, the calculated valorization scores were classified to generate the final valorization map, allowing the identification of samples with comparatively higher valorization potential. Each step generated a defined output that served as the input for the final step, providing a transparent and reproducible workflow from the original VOC dataset to the final valorization assessment and map.

Dataset description and preprocessing

This study used secondary dataset. The data were obtained from open-access dataset available at https://doi.org/10.5281/zenodo.16410077 [2]. Carriero et al. (2025) synthesized published data on VOCs emissions in grapevine berries at different developmental stages (BBCH), from fruit set to harvest. The original database compiled VOC profiles from various cultivars of red and white grapevine (Vitis vinifera L.) cultivated across nine countries. It included chemical profiles obtained from different grapevine tissues including skin, pulp, and must and characterized them in terms of their free and bound VOC fractions [1].

In this study, analysis was performed through selection of key parameters to develop the valorization framework, including VOCs emission, grapevine cultivar, berry type (red, white), berry part (skin, pulp, must), chemical forms (free and bound), and BBCH. From the original dataset of 69 berry samples, 13 were excluded, because of missing information regarding tissue source or VOCs chemical forms. Consequently, 56 samples retained and assigned new Ids as detailed in Table S1.

To visualize the VOC fingerprint of the grapevine berry across the studied dataset, relative abundance (R) was calculated as the ratio of the emission intensity of each subclass to the total VOC emission of the corresponding sample, following the approach described by Wang et. al． [13].

$$C_{subclass}=\sum _{i=1}^n{C}_i$$ (1)

In Eq. 1, the C _subclass represents the total emission of all compounds belonging to a specific VOC subclass within the assigned sample, calculated as the sum of the VOC emission of the individual compounds (C_i) that make up that subclass. Here, n denotes the total number of compounds assigned to that subclass.

$$R_{subclass}=\frac{C_{subclass}}{C_{total}}\times 100$$ (2)

Eq. 2 shows the formula used to calculate the R of each subclass. According to this formula, R _subclass was obtained by normalizing C _subclass to the total VOC emission of the given sample (C _total). The C _total was calculated as the sum of emissions of all detected VOC across all subclasses in the assigned sample. R _subclass is a normalized measure of the proportional contribution of each chemical subclass to the total VOC composition within an individual sample. The VOC fingerprint of the grapevine berry across the studied dataset is presented in Fig. 1. It is worth mentioning that since the VOC emission intensities spanned several orders of magnitude, plotted values were in a logarithmic (log₁₀) scale to enable clearer visualization. Since emission values below 1 µg kg⁻¹ produces negative log₁₀ outputs, making interpretation complicated, emission intensities were multiplied by 10 prior to logarithmic transformation, to ensure all values fell within a positive range. The adjusted values were then used for consistent scaling in visualization.

Fig. 1.

The VOC relative abundance within and across the 56 samples of the studied dataset. The x-axis represents the sample Ids (n=56), the y-axis displays the VOC subclasses, and the color intensity denotes the relative abundance (R) of each subclass .

Among the 13 VOC subclasses specified within the dataset, seven major subclasses dominated the total profile with over 99.7% of the total VOC emissions. Carboxylic acids were the most abundant subclass, accounting for 38.9%, followed by alcohols at 31.1%. Other major contributors included aldehydes (9.9%), terpenoids (9.5%), and esters (8.9%), while Ketones (1.3%) and phenols (0.1%) contributed slightly. The remaining six subclasses collectively contributed less than 0.3% of the total VOC profile. These findings are consistent with previous reports on the relative abundance of VOC subclasses in grapevine berries. Lee et al [14] characterized grapevine VOC during three ripening stages including green, pink and fully mature. A total of 28 VOC, including esters, alcohols, terpenoids, and carbonyl compounds, were identified, with their concentrations varying significantly across ripening stages. In early ripening, terpenoids were the major VOC subclass found. As ripening progressed, aldehydes gradually increased and surpassed the early dominant group. At the final stage, esters characterized as the most abundant VOC subclass [14]. Kaya et al [15] investigated two grape varieties throughout six different ripening stages. They identified terpenes, aldehydes, fatty acids, and alcohols as the dominant VOC subclasses. They reported that the concentration of most VOCs increased as ripening progress from the begin of berry touch to ripe berries for harvest stages [15]. It is apparent from the heatmap presented in Fig. 1 that seven subclasses were the dominant chemical contributors across this dataset. Hence, they were selected for further downstream valorization mapping.

This dataset reported various parameters which may affect the VOC emission intensity. These parameters include grapevine berry part, grapevine berry type, VOCs chemical form, grapevine variety, and BBCH. To identify key influencing features, decision tree analysis was conducted. A single decision tree extracted from the Random Forest model, with its depth limited to 3 for readability was used to quantify the contribution of each feature to the emission intensity (Fig. 2).

Fig. 2.

A representative decision tree extracted from the random forest model (depth = 3) illustrating the relative contribution of each feature to VOC emission intensity.

As shown in Fig. 2, the most effective feature is BBCH with a corresponding score of 0.24, suggesting that it is the primary driver of VOC emission intensity in the investigated dataset. The second feature is the free chemical form of VOCs with an approximate score of 0.21. The free fraction shows the directly accessible VOCs profile [16]. Another feature with lower score (approximately 0.14) is the grapevine variety, specifically Moristel. The skin part of the berry is the fourth feature which influences the VOC profile but with a lower score (0.13). Despite the fact that skin is a rich source of VOCs [17], it is assigned a lower score due to its low mass contribution to the total berry weight. The white berry type is another feature with a corresponding score of 0.12. Pulp also contributes, though to a lesser extent than skin (score of 0.06). These results are align with the Carriero and coworkers study which listed grapevine variety, BBCH, and grapevine tissue source as the notable parameters [1].

The free-fraction distributions of the seven major VOC subclasses across the 56 berry samples are shown in Fig. 3a. Because recovery feasibility depends jointly on extractability (free fraction) and quantity (relative abundance), only samples that combine a high free fraction with a high relative abundance constitute viable recovery targets; samples with predominantly low free-form emission have limited extractable content and contribute little to downstream recovery [16]. A recurring pattern across most subclasses is that many samples reached a free fraction of 1 yet carried low relative abundances, meaning their free-form VOCs are extractable in principle but not economically recoverable. The clearest exceptions were ketones in sample 36 (relative abundance 0.96, free fraction of approximately 1) and aldehydes in samples 5 and 6 (relative abundance >0.99, free fraction 1), which emerged as the strongest candidates for targeted extraction. These samples were all sourced from grape skin at advanced phenological stages (BBCH 81 for sample 36; BBCH ≥81 for samples 5 and 6; Fig. 3b) and corresponded to the Shiraz and Chardonnay cultivars (Fig. 3c;). In contrast, the most abundant terpenoids (samples 2, 25, 39; relative abundance >0.97) and the highest-abundance ester (sample 3, Pinot noir; 0.90) occurred predominantly in the glycosidically bound form, so their direct extractability is limited despite high availability and would require releasing from the bound fraction. Carboxylic acids, alcohols, and esters generally showed only moderate relative abundances (e.g., 0.71 and 0.68 for acids in samples 34 and 35; 0.88 for alcohols in the Aglianico sample 50), rendering industrial extraction economically challenging, whereas phenols exhibited uniformly very low relative abundances (max 0.17) and are therefore not economically viable for recovery. Overall, Fig. 3 indicates that genuine VOC recovery potential is concentrated in a small subset of skin-derived, late-stage samples within the ketone and aldehyde subclasses.

Fig. 3.

(a) Free fraction patterns of seven major VOC subclasses across 56 berry samples; (b) Distribution of 56 grape samples by BBCH, Individual data points represent samples harvested at a single, specific phenological stage, while vertical bars indicate samples characterized by a range of developmental stages; (c) Distribution of 56 grape samples by variety, the number of each dot shows the sample Ids, while the corresponding cultivar names are provided as y-axis labels.

Framework development

From a valorization standpoint, samples combining high emission intensity with elevated free fractions, represent the most efficient targets for recovery and downstream conversion into food, flavour, bioactive ingredients, cosmetics, and pharmaceutical [18]. Conversely, compounds with low free fractions may require targeted pretreatments such as enzymatic or hydrolytic approaches or physical release methods like ultrasound to unlock their valorization potential, with the choice of method depending on the target compound subclass and acceptable processing conditions [19]. However, to ensure robust decision-making, it is essential to integrate additional factors into the selection of the optimal valorization pathway. Considering the diversity index of the target sample and the market performance of the resulting extract are crucial because without this information, the downstream valorization pathways may target the wrong samples.

Diversity index

Addressing diversity index is critical to fully harness the valorization framework and establish it as a viable, sustainable alternative to traditional waste disposal strategies. Hence, the Shannon diversity index (H) was calculated to quantify the chemical heterogeneity of VOCs in each individual sample, following the approach of Izsák and Papp [20], computed according to the Eq. 3.

$$H=-\sum _{i=1}^n{p}_{i}Ln{p}_i$$ (3)

Where p_i was the proportion of each compound relative to the total emission of the sample and H showed the Shannon diversity index. The calculated total VOC emission and H of each sample are reported in Table S2 and they are visualized in Fig. 4.

Fig. 4.

Characterization of VOC emission profiles and Shannon diversity index across 56 grape berry samples: (a) Distribution of logarithmic VOC emission (light blue) and Shannon diversity index (light orange); (b) Distribution of Shannon diversity index categorized by berry part: skin (light blue), skin & pulp (light orange), and must (rose); (c) Relative abundance of major VOC subclasses across berry parts; (d) Distribution of Shannon diversity index categorized by berry type: red (rose) and white (light yellow).

Across the 56 samples, log-scaled emission intensity and H (Fig. 4a) were moderately negatively correlated (Pearson r = −0.5): the most diverse samples (31 and 32, H = 3.50) drew balanced contributions from many subclasses, whereas the least diverse (sample 36 at 0.27, dominated by ketones; samples 5 and 6, dominated by aldehydes) were chemically specialized yet still carried medium-to-high total emissions, making this low-diversity, high-emission combination the most attractive for targeted recovery owing to their lower analytical complexity during extraction and purification [21]. Analysis by tissue type showed that skin (Fig. 4, Fig. 4) held the widest diversity range (0.27–3.50) with a high abundance of terpenoids. In contrast, must samples exhibited the highest specialization (H < 2.00 for 91%), dominated by carboxylic acids and alcohols. The mixture of skin and pulp fraction showed intermediate diversity; aligned with previous findings on skin-enriched terpenoid biosynthesis [22]. However, neither tissue source (one-way ANOVA, p = 0.242) nor berry type (red vs. white; t-test, p = 0.539) differed significantly, red berries (Fig. 4d) spanned a broader H range than white, but the difference was not statistically supported, partly owing to unequal group sizes. Therefore, biochemical factors, rather than botanical categories, are the primary drivers of VOC diversity. Hence, diversity index was selected as one of the core indices for the valorization framework.

Market evaluation index

Inspired by multi-product biorefinery approaches for grape waste valorization [23], and sustainability-oriented extraction assessment frameworks [24], a composite market value metric was designated and called as Market Potential Index (MPI). It is defined according to Eq. 4:

$$MP{I}_i=MVB{S}_i+FD{S}_i+PD{P}_i$$ (4)

Here, $MP{I}_i$ is the market potential index, and $MVB{S}_i$, $FD{S}_i$, and $PD{P}_i$ are market value base score, functional demand score, and purification difficulty penalty, respectively. The integrated market analysis facilitated the selection of VOCs that combine an optimal balance between market value, functional properties, and recovery feasibility.

The MVBS shows the relative economic contribution of each chemical subclass. To calculate this metric, first we calculated the market value (MV) according to Eq. 5, by multiplying the compound’s emission intensity $C_i$ by its corresponding market price ($P_i$, USD kg⁻¹). For each VOC subclass at each individual sample, the compound with the highest emission intensity within that subclass was selected as the representative lead of the group [25]. Then the normalization of the market value was performed by dividing the calculated MV by the maximum observed market value (MV_max) across all samples, as defined in Eq. 6

$$M{V}_i=C_i\times P_i$$ (5)

$$MVB{S}_i=\frac{M{V}_i}{M{V}_{max}}\times 100$$ (6)

This normalization allows direct comparison of VOC subclasses in terms of their relative market relevance. Calculated values are summarized in Table S3.

The FDS quantified the relevance of each VOC subclass across diverse applications by integrating literature-derived insights [24] and insights generated from annotated database entries [26]. To calculate this index, each compound's market demand was identified using PubChem's compound classification data. This dataset documents reported industrial and commercial applications of compounds across various categories such as fragrance and perfumery. Each functional class was then assigned a market demand score on a scale of 1.0 to 5.0 based on documented industry-scale significance, as summarized in Table 1. Then the average of these assigned scores was calculated and reported as the FDS score to provide a standardized metric for comparing functional relevance across chemically diverse VOC subclasses.

Table 1.

Detailed FDS scores by application category with supporting justifications.

Class	Market value (USD)	CAGR (%)	Score	Score Justification
Medicinal	375.6 billion (2032)	7.2	5.0	•High market value •Highest CAGR
Cosmetics & skincare	326.0 billion (2031)	–	4.0	•High market value
Fragrance & perfumery	30.6 billion (2023)	5.4	3.5	•Medium market value •Higher market demand
Flavouring & food additives			3.0	•Medium market value •Lower market demand
Solvents & intermediates	430.0 million (2030)	3.8	2.5	•Low market value
Industrial	–	–	2.0	•Low market value •Moderate market demand
Research & analytical uses	–	–	1.0	•Laboratory-only demand

The global medicinal plants market was valued at USD 201.0 billion in 2023 and is expected to reach to a value of USD 375.6 billion by 2032, reflecting a CAGR of 7.2 % [27], assigning a the highest score of 5.0. The cosmetic and skincare sector is projected to reach over USD 326.0 billion by 2031 [28]. Since it was assigned the second score of 4.0. Fragrance and perfumery received the third score, 3.5, reflecting its extremely high global commercial demand. The fourth place assigned to flavouring and food additives with a score of 3.0. The global fragrance and flavour market was worth at approximately USD 30.6 billion in 2023, with an anticipated CAGR of 5.4% from 2024 to 2030 [29]. Since, the fragrance and flavour market produced 0.8 million tons material in 2025 of which around 0.6 million tons were attributed to the fragrance segment [29], a higher score was assigned to fragrance segment. In the solvent and intermediates class, ethyl acetate emerged as the dominant chemical by application volume, making it the natural reference point for score assignment. The global market for bio-ethyl acetate was valued at USD 165.0 million, with an estimated CAGR of 3.8 %, and is projected to reach around USD 430.0 million by 2030 [30]. Hence, this functional class was assigned a score of 2.5. Since the compounds in the industrial group serve steady but narrowly defined manufacturing and materials applications with moderate overall market volume, they were assigned a score of 2.0. Research and analytical uses received the lowest score, 1.0, reflecting their limited commercial scale and laboratory-only demand. Finally, the FDS for each compound was computed as the arithmetic means of the scores corresponding to all functional classes in which it was identified, ensuring that multi-functional compounds are rewarded proportionally for broader market relevance. For instance, the hexanol detected in sample 1, has various applications mapped to five relevant functional classes including medicinal (5.0), fragrance (3.5), flavouring (3.0), solvent (2.5), and industrial applications (2.0). Hence, the hexanol received an FDS score of 3.2,representing its overall application relevance across the identified functional classes.

The PDP shows the purification difficulty penalty. It was applied as a corrective adjustment to account for physicochemical properties that impose measurable downstream processing burdens. Hence it incorporated factors that complicate extraction of VOCs from grapevine berry, including high volatility of light VOCs, and thermal sensitivity of the VOCs [24]. In this study, relative abundance was not incorporated as a penalty parameter in the PDP calculation. The reported VOC emission range in the dataset spans from 0.2 to 3,148,451 µg kg⁻¹, representing an extraordinarily wide concentration range. However, even the lowest detected concentration in this dataset (0.2 µg kg⁻¹) exceeds the literature-reported detection limit threshold for grape berry VOCs (0.02 µg kg⁻¹) [31], below which dedicated pre-concentration steps become analytically necessary [32]. Since all compounds in the dataset were detectable above this critical threshold, none met the condition for which a low abundance penalty would be warranted, and therefore this parameter was not considered in the PDP calculations. The assigned scores for the two incorporated factors including volatility and thermal sensitivity are reported in Table 2.

Table 2.

The physicochemical properties considered for calculating the PDP and their assigned scores.

Physicochemical property	Score
Volatility	–1.0
Thermal sensitivity	–1.0

The recovery of light VOCs is complicated due to their high volatility and pronounced tendency for evaporation lost. Hence it is mandatory to use specialized low-temperature condensation or cryogenic tapping strategies. These techniques complicate the routine laboratory and pilot-scale extraction procedures. Moreover, the high energy demand and potential physical losses of the VOCs during such intricate processes result in the reduction of the overall yield [33]. Thermal sensitivity imposes a comparable burden on compound recovery yield through chemical degradation during processing. Terpenes, aldehydes, and other oxygenated compounds are prone to degradation, isomerization, or polymerization under elevated temperatures. Hence, gentle extraction and purification methods such as ultrasound-assisted and supercritical fluid extraction are required to operate at low temperatures, preserving the VOC bioactivity and chemical integrity, though their implementation is costly [34,35]. Since high volatility and thermal sensitivity introduce additional process constraints that, while significant, can be mitigated through selection of appropriate separation techniques without fundamentally compromising overall recovery potential, a similar score of -1.0 was assigned to them.

Compounds with a reported vapour pressure exceeding 0.1 hPa at 25°C were classified as volatile and were assigned a volatility penalty of -1. This threshold aligns with the definition established under the EU VOC Directive (1999/13/EC) [36], which identifies compounds above this vapour pressure as volatile organic compounds subject to significant evaporative loss during industrial processing. In the case that the vapour pressure information was unavailable, the boiling point was used as the reference and the VOCs with boiling points lower than 220°C was considered as volatile [37]. Thermal sensitivity score was assigned using a literature-based kinetic degradation framework that categorizes VOCs based on their degradation kinetic constant and precursor hydrolysis kinetic constant [38]. According to this framework, all alcohol and carboxylic acids in the studied dataset were assigned to a thermal sensitivity score of 0, with a score of -1.0 for all aldehydes. All compounds classified in the ester subclass were assigned a score of 0, except methyl geranate, methyl vanillate, and ethyl vanillate, which were assigned a score of -1.0. From ketone subclass, six compounds including 1-octen-3-one, acetovanillone, zingerone, trans-geraniol, vanillyl methyl ketone, B-Inonone, were assigned a score of -1.0. From the terpenoids subclass, all compounds were assigned a score of -1.0 except two: 1,8-cineole and 6-methyl-5-hepten-2-ol. Most compounds from the phenol subclass were considered thermally stable under typical biorefinery processing conditions, and no penalty was applied.

In this dataset, terpenoids (9.5% for relative abundance) recorded the highest cumulative MVBS, reaching 225.0, which was two to three times higher than the other major subclasses, confirming their exceptionally strong economic baseline, and the highest cumulative FDS of 180.6 and cumulative PDP of -77.0. This dominance can be attributed to several converging factors: high consumer demand for floral and citrus sensory profiles [39], widespread cross-sector utilization [40], and relatively favorable extraction feasibility due to moderate polarity and volatility [41]. Moreover, for terpenoids, high commercial value can be achieved by extracting minor quantities, due to their low sensory thresholds. Additionally, their relative chemical stability reduces formulation challenges and enhances long-term market reliability [39]. Carboxylic acids, the most abundant VOC subclass (38.9% for relative abundance), displayed a moderate cumulative MVBS of 8.6 and the lowest cumulative FDS of 92.4 among the seven major VOC subclasses. The most abundant identified acid in this dataset included short-chain carboxylic acids (e.g., acetic, propanoic, butanoic), medium-chain fatty acids (e.g., hexanoic, octanoic, decanoic), long-chain fatty acids (e.g., palmitic, oleic, linoleic), and phenolic acid derivatives, which were available bulk chemicals. Since these compounds can be produced economically through large-scale petrochemical [42] or microbial fermentation processes [43], their functional demand in grape berry-derived tissues valorization system was comparatively limited, despite their high abundance. Esters with a relative abundance of 8.9%, recorded a low cumulative MVBS of 2.0 and a high cumulative value of 141.5 for FDS, together with a cumulative PDP score of -44. These values indicated that, despite their low base market prices, esters were highly demanded across multiple sectors. Their strong fruity and floral aroma characteristics, make them highly attractive for flavour and fragrance applications [44]. Alcohol, with a relative abundance of 31.1%, showed a similar trend. They recorded a low cumulative MVBS of 1.3, a high cumulative value of 156.6 for FDS, and a cumulative PDP score of –39. These inexpensive bulk chemicals are key intermediates in aroma formation and fermentation-derived flavour chemistry, with a significant impact on the sensory properties [45]. Aldehydes accounted for a 9.9% relative abundance, yielding cumulative values of 1.1, 116.0, and –74.0 for MVBS, FDS, and PDP, respectively. Two primary factors explain their low market value despite moderate functional demand. First, petrochemical or enzymatic synthesis of many short-chain aldehydes at industrial scale creates a saturated competitive market that suppresses prices [46,47]. Second, high chemical reactivity of aldehydes (e.g. oxidation, reduction, and polymerization reactions) causes severe formulation instability [48,49], which is reflected in the extremely recorded PDP score of –74.0 for this subclass, resulting in their restricted premium applications. The minimum relative abundance within the dataset represented by ketones and phenols with values of 1.3% and 0.1%, respectively. For these two subclasses, the cumulative MVBS values were negligible (0.2 and 0.0, respectively). However, ketones maintained a moderate cumulative score of 106.3, while Phenols recorded a relatively low score of 88.2 for FDS. These values reflected their less pronounced role in the VOC aroma profile. The most interesting finding was that abundance alone does not determine valorization potential.

Valorization mapping framework

The defined indices including E, F, H, and MPI were integrated into a four-route valorization framework to translate the analytical findings of this study into practical valorization routes. These four valorization strategic routes included targeted high-value extraction (route 1), functional bioactive recovery (route 2), commodity-level industrial processing (route 3), and secondary bioprocessing and upgrading (route 4). Route 1 was characterized by high E, high F, low H, and high MPI, capturing samples containing VOCs that serve as premium candidates for high-tier market sectors like aroma and fragrance. Route 2 was defined by moderate E, moderate F, low H, and moderate MPI. Samples that are ideal candidates for the recovery of complex mixtures utilized primarily for their nutritional, antioxidant, or biological functions were categorized in this group. Route 3 was characterized by moderate E, moderate F, moderate H, and low MPI, targeting high-abundance samples where large-scale recovery is feasible, despite relatively low market values. Route 4 was characterized by low E, low F, high H, and low MPI. Samples which serve as efficient feedstocks for microbial conversion, fermentation, or chemical upgrading processes were categorized in this group. This mapping framework establishes a waste classification model that categorizes samples based on their recovery feasibility and commercial viability. Hence, it addresses the limitations of traditional models that fail to capture the system complexity, thereby providing a structured strategy for sustainable waste management.

The specific categorization criteria and numerical ranges for the four integrated indices are detailed in Table 3.The thresholds for E [31,50] and F [51] were defined based on the ranges reported in the literature. Since the diversity index and MPI are dependent variables derived from the present dataset, their classification thresholds were determined empirically based on the internal data distribution rather than external literature values. For the diversity index, quartile-based classification [52] was applied to the raw values, dividing the observed range into four equal-frequency categories representing low, low to moderate, moderate to high, and high compound diversity. The MPI values exhibited a highly skewed distribution where most values fell below 3.0 while extreme values reached 102.0. Hence, quartile classification was selected and applied to log-transformed data. This approach ensured meaningful and balanced categorization across the full dataset, reflecting the natural variation in the dataset rather than arbitrary equal-interval divisions.

Table 3.

Categorization criteria and numerical ranges for the integrated valorization indices including E, F, H, and MPI.

Group	Threshold	Justification	Ref
E
Very low	< 10 (µg kg−1)	Below typical detection significance threshold	[31,50]
Low	10–100 (µg kg−1)	Trace level, preconcentration required
Moderate	100-1000 (µg kg−1)	Recoverable with standard techniques
High	> 1000 (µg kg−1)	Highly abundant, primary recovery target
F
Very low	< 0.25	Predominantly bound, requires hydrolysis	[51]
Low	0.25–0.50	Mostly bound with some free fraction
Moderate	0.50–0.75	Balanced bound and free fractions
High	0.75–1.00	Predominantly free, directly extractable
H
Very low	0.25–1.73	Very low compound diversity	[52]
Low	1.73–2.22	Low compound diversity
Moderate	2.22–2.73	Moderate diversity
High	2.74–3.50	High compound diversity
Log (MPI+1)
Very low	0.00–0.40	Low commercial priority	[52]
Low	0.40–0.48	Below average priority
Moderate	0.48–0.54	Above average priority
High	0.54–2.01	High commercial priority

The valorization route assignment for each VOC subclass was determined based on the combined profile of E, MPI, H, and F, as summarized in Table 4. In cases where index values deviated partially from the ideal route profile, assignment was based on the domain-specific knowledge of each compound subclass and its established industrial processing characteristics. Evaluation of the optimum valorization pathway for each subclass was performed by selecting one representative exhibiting the peak emission intensity, to ensure the framework was tested against the most resource-concentrated streams. By isolating these high-emission markers, the framework simplified the complexity of the multi-component volatile profiles into a discrete set of target analytes, facilitating more precise modeling of the extraction efficiencies and valorization potentials within the circular bioeconomy system.

Table 4.

Integrated VOC valorization routing framework: assignment of the seven major VOC subclasses to sector-specific processing pathways based on E, F, H, and MPI. The MPI stands for log (MPI+1).

VOC subclass	Representative	Valorization indicators	Valorization pathway classification	Technical justification
Terpenoids	Geranyl Acetone	•Id: 10 •E: 93020 •F: 1 •H: 1.5 •MPI: 0.62	Targeted high value extraction (Route 1)	•E: High •MPI: High •H: Low •F: High
Ketones	Gamma-Butyrolactone	•Id: 41 •E: 61174 •F: 0 •H: 2.2 •MPI: 0.36	Secondary bioprocessing upgrading (4)	•E: High •MPI: Very low •H: Low •F: Very low
Phenols	4-Vinylguaiacol	•Id: 16 •E: 3017 •F: 0 •H: 2.8 •MPI: 0.40	Secondary bioprocessing upgrading (4)	•E: High •MPI: low •H: High •F: Very low
Carboxylic acids	Acetic Acid	•Id: 41 •E: 3148451 •F: 0 •H: 2.2 •MPI: 0.42	Secondary bioprocessing upgrading (4)	•E: High •MPI: Low •H: Low •F: Very low
Esters	Ethyl Acetate	•Id: 41 •E: 347168 •F: 0 •H: 2.2 •MPI: 0.40	Secondary bioprocessing upgrading (4)	•E: High •MPI: Low •H: Low •F: Very low
Alcohols	Isoamyl Alcohol	•Id: 41 •E: 1818850 •F: 0 •H: 2.2 •MPI: 0.49	Functional bioactive recovery (2)	•E: High •MPI: Moderate •H: Low •F: Very low
Aldehydes	Acetaldehyde	•Id: 53 •E: 284537 •F: 0 •H: 1.7 •MPI: 0.46	Secondary bioprocessing upgrading (4)	•E: High •MPI: Low •H: Very low •F: Very low
Carboxylic acids	Acetic Acid	•Id: 47 •E: 28 •F: 0.7 •H: 2.7 •MPI: 0.40	Commodity level industrial processing (3)	•E: Low •MPI: Low •H: Moderate •F: Moderate
Terpenoids	Linalool	•Id: 43 •E: 257 •F: 0 •H: 2.5 •MPI: 0.40	Commodity level industrial processing (3)	•E: Moderate •MPI: Low •H: Moderate •F: Very low
Ester	Methyl vanillate	•Id: 22 •E: 239 •F: 0.5 •H: 2.3 •MPI: 0.48	Commodity level industrial processing (3)	•E: Moderate •MPI: Low •H: Moderate •F: Moderate
Aldehyde	Vanillin	•Id: 20 •E: 31 •F: 1 •H: 2.4 •MPI: 0.48	Secondary bioprocessing upgrading (4)	•E: Low •MPI: Low •H: Moderate •F: High
Ester	Butyl acetate	•Id: 23 •E: 119 •F: 0.57 •H: 2.3 •MPI: 0.44	Commodity level industrial processing (3)	•E: Moderate •MPI: Low •H: Moderate •F: Moderate

*Id: show the sample Id.

For instance, route 1, targeted high-value extraction, focused on samples that combine high E, high MPI, high F, low H. Among the VOC subclasses listed in Table 4, terpenoids from sample 10 was most closely fitted with these criteria, achieving a MPI score of 0.62, indicating broad market leverage. These terpenoids served as ideal candidates for high-value applications across flavour, fragrance, cosmetics, and nutraceutical sectors [22]. A moderate diversity index (1.5), alongside the presence of target chemicals in their free form facilitated the extraction of terpenoids from this sample. Moreover the low sensory thresholds of these compounds resulted in substantial economic returns for minor recovered volumes [39]. Assignment of acetic acid to route 3 in sample 47 and route 4 in sample 41 provided the clearest demonstration in the dataset of the emission-accessibility paradox inherent in VOC valorization assessment. Interestingly, acetic acid in samples 41 was directed to a lower valorization route (route 4: secondary bioprocessing) despite exhibiting higher emission intensity (3148451 µg kg⁻¹) than sample 47 (28 µg kg⁻¹), which was assigned to a higher valorization route (Route 3: commodity-level industrial processing). This unexpected result can be explained by the tightly bound nature of acetic acid in sample 41. Consequently, only a tiny free fraction is accessible for direct recovery, while extracting the bound fraction requires costly pretreatments that cannot be justified economically. In contrast, sample 47 enabled commodity-level processing of a directly accessible and compositionally diverse fraction. Its balanced free fraction and diversity metrices offset its lower emission intensity, making direct processing feasible. Since MPI was identically low in both samples 47 and 41, confirming consistent commodity-grade commercial value of acetic acid regardless of sample origin, the route divergence between these two samples was driven entirely by differences in free fraction and diversity index. This finding confirmed free fraction as the primary determinant of route 3 versus route 4 boundary for commodity-grade compounds and underscored the inadequacy of emission intensity alone as a valorization indicator in biorefinery screening frameworks.

The route assignments in Table 4 revealed that the VOC subclass alone was insufficient to determine the optimal valorization strategy. Instead, commercial and technical feasibility was determined by specific combination of E, E, H, and MPI values associated with each individual sample. This distinction were illustrated by terpenoid subclass, which spans Route 1 and Route 3, and the acetic acid and ethyl acetate compounds, each of which appear under both Route 3 and Route 4. In all cross-route cases, F emerged as the critical differentiating parameter between commodity industrial processing (route 3) and secondary bioprocessing (route 4). However, MPI served as the primary determinant of the boundary between high-value targeted extraction (route 1) and the other three lower-value routes. These findings suggested that the four-index valorization framework provides meaningful discrimination between samples of the same VOC subclass. Beyond basic compound classification, this framework established a systematic, data-driven basis for biorefinery route selection.

Method validation

Sensitivity analysis of MPI weighting schemes

To evaluate the stability of compounds rankings under variations in sub-index weights, a sensitivity analysis across four alternative scenarios was conducted [53]. Table 5 summarized these four weighting scenarios. While the original weighting scheme was grounded in market-based and physicochemical reasoning as described in the preceding sections, it is acknowledged that alternative weighting choices could reasonably be made depending on the priorities of a given biorefinery context. This analysis tested whether the overall ranking of compounds changes substantially by applying different weighting schemes. The stable ranking across multiple scenarios provided strong evidence that the MPI output reflected the underlying data rather than the specific weight assumptions, thereby demonstrating the robustness of the proposed index.

Table 5.

Summary of four weighting scenarios used to evaluate the robustness of the MPI index.

Scenario	MVBS weight	FDS weight	PDP weight	Justification
Scenario 1: Equal weights	1.0	1.0	1.0	Neutral
Scenario 2: Market-focused	0.6	0.2	0.2	Prioritize emission value
Scenario 3: Function-focused	0.2	0.6	0.2	Prioritize functional demand
Scenario 4: Purification-focused	0.2	0.2	0.6	Prioritize recovery feasibility

To assess the robustness of the MPI indices [53], four alternative weighting scenarios (Table 5) were applied. Scenario 1 evaluated an equal weighting scheme (MVBS: 1.0, FDS: 1.0, PDP: 1.0), scenario 2 focused on market value scheme (MVBS: 0.6, FDS: 0.2, PDP: 0.2), scenario 3 was a function-focused scheme (MVBS: 0.2, FDS: 0.6, PDP: 0.2), and scenario 4 prioritized purification efficiency (MVBS: 0.2, FDS: 0.2, PDP: 0.6). Across all four scenarios, most compounds showed relatively stable MPI rank placements. This overall consistency demonstrated that the prioritization framework was robust against moderate variations in component weighting. Nevertheless, marked rank shifts within a minor subset of compounds revealed that a few selected compounds remain sensitive to weight allocation. A key norisoprenoid terpenoid, 3-Oxo-alpha-ionol, is a representative example, detected in two distinct samples and yielded the most extreme MPI values across all evaluated scenarios. This compound reached MPI values of 21.6 and 60.4, under the function-focused scheme (FDS: 0.6) and the market value-focused scheme (MVBS: 0.6), respectively. This compound achieved its highest potential (MPI: 102.0) under the equal weighting scheme (MVBS: 1.0, FDS: 1.0, PDP: 1.0). Conversely, its MPI score dropped substantially under the purification-focused scheme (PDP: 0.6), where its high purification difficulty penalties exerted the greatest influence. This observed pattern confirmed that the rank sensitivity of this compound was associated with its high MVBS and FDS values. Higher weighting of either or both components increased its MPI score. Therefore, the relative rank was affected. Importantly, this rank instability was observed for only a small number of observations within the dataset. However, the majority of the 285 observations maintained consistent relative rankings across all four weighting scenarios. These findings highlighted the overall robustness of the MPI index at the dataset level. The limited rank sensitivity was confined to a small number of chemically distinctive compounds with particularly exceptional market value and functional demand. Hence, the sensitivity observed did not compromise the validity of the framework, rather highlighted the importance of transparent weight selection that reflects the specific priorities of the biorefinery context.

Spearman rank correlation helps to determine how well the rankings of units in a composite index are preserved when various weights are used. This helps in understanding the robustness of the index's output to changes in its construction [53]. The Spearman rank correlation coefficients between all pairs of weighting scenarios are presented in Table 6.

Table 6.

Pairwise Spearman rank correlation matrix of compound rankings across four MPI weighting scenarios including Scenario 1: Equal weight, Scenario 2: Market-focused, Scenario 3: Function-focused, and Scenario 4: Purification-focused.

	Scenario 1	Scenario 2	Scenario 3	Scenario 4
Scenario 1	1.00	0.99	0.88	0.99
Scenario 2	0.99	1.00	0.89	0.98
Scenario 3	0.88	0.89	1.00	0.83
Scenario 4	0.99	0.98	0.83	1.00

All pairwise correlations exceeded 0.83. These results confirmed strong to very strong rank agreement across all evaluated weighting schemes. The highest correlation observed between the equal weighting scheme (MVBS: 1.0, FDS: 1.0, PDP: 1.0) and the market value-focused scheme (MVBS: 0.6, FDS: 0.2, PDP: 0.2), with a coefficient of 0.99, confirming nearly identical compound rankings. Similarly, a strong correlation of 0.98 observed between the market value-focused and purification-focused schemes (MVBS: 0.2, FDS: 0.2, PDP: 0.6). This strong correlation demonstrated high ranking consistency between these contrasting weighting approaches. The most contrasting weighting assumptions in the analysis was observed for the function-focused scheme (MVBS: 0.2, FDS: 0.6, PDP: 0.2) and the purification-focused scheme (MVBS: 0.2, FDS: 0.2, PDP: 0.6), with the lowest correlation value of 0.83. Nevertheless, this minimum correlation of 0.83 reflected strong rank agreement, proving that compound prioritization remained largely consistent regardless of the specific weighting assumptions applied. These results proved that the MPI framework is robust. Moreover, this framework outputs depends primarily on the physicochemical and market metrices rather than sensitivity to component weights.

Comparison with conventional models

The four-route valorization framework developed in this study advances beyond existing biomass waste valorization approaches in several important aspects. While recent tools such as the Biomass Valorization Recommender [54] and hierarchical biomass valorization frameworks [55] provide systematic pathway selection at the feedstock and technology level, they do not operate at the compound resolution required for targeted VOC extraction process design. Multi-criteria decision analysis approaches, while structurally similar in their use of weighted scoring criteria, predominantly rely on expert elicitation for weight assignment and rarely provide quantitative robustness validation [54], limitations that the present framework addresses through market and physicochemical data-driven weight justification and sensitivity analysis demonstrating Spearman rank correlations consistently exceeding 0.83 across all weighting scenarios. Compared to techno-economic analysis [56], which provides detailed economic validation but requires extensive process engineering data unavailable at early screening stages, the present framework serves as a data-efficient pre-screening tool that identifies high-priority valorization targets using only compound-level emission, market, diversity, and physicochemical accessibility data. The framework's proposed route assignments are consistent with processing strategies validated in biorefinery literature [57], such as sequential extraction for functional bioactive recovery and integrated cascade processing for commodity-level industrial applications. These findings confirm the practical relevance of the framework outputs for making informed biorefinery design decisions.

Limitations

Several limitations should be acknowledged when interpreting this study findings. One limitation related to the dataset used which was synthesized from previously published literature rather than generated through new experimental measurements. Hence, the emission intensity, free fraction, and diversity values reflect the conditions and methodologies of the original studies rather than a standardized single experimental protocol. The prices adopted in this study were derived from Sigma-Aldrich reagent-grade pricing as a proxy for market value calculations. These laboratory-scale prices may not accurately reflect actual bulk industrial market prices, which are typically lower and subject to supply-demand fluctuations. Third, the valorization route assignment framework was based on discrete categorical thresholds (low, moderate, high) for each index. Borderline samples that did not perfectly match any single route required a nearest-match assignment approach. This systematic approach introduces an additional layer of interpretive judgement. Finally, the proposed framework has not yet been validated against pilot-scale or industrial extraction trials; as such, the MPI and valorization route assignments should be regarded as a data-driven screening and prioritization tool rather than a definitive substitute for detailed process-level economic and environmental validation. Future work should focus on experimental validation of the proposed valorization routes and integration of the framework with techno-economic and life cycle assessment approaches to confirm its practical applicability in real biorefinery settings.

Ethics statements

Not applicable.

CRediT author statement

Hajar Rastegari: Conceptualization, Methodology, Formal analysis, Validation, Visualization, Writing-original draft, Writing-review and editing. Ramin Khezri: Conceptualization, Writing-review & editing. Hassan Nasrollahzadeh Saravi: Writing-review & editing. Romi Fadillah Rahmat: Writing-review and editing. Farhad Nadi: Methodology, Validation, Writing-review & editing.

AI usage disclosure

During the preparation of this work the authors used Google Gemini, and Claude for English language refinement, grammar correction, and improvement of scientific readability.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

The data were obtained from previously published open-access dataset accessed via Zenodo under DOI: 10.5281/zenodo.16410077 and described in Data in Brief (Carriero et al., 2025a).