The research team, led by Haoran Liu of Wuhan Polytechnic University, focused on sunflower oil as their starting material. Sunflower oil is an attractive feedstock because it is more than 96 percent digestible and rich in tocopherols, phytosterols, and trace elements. The researchers used immobilized Candida antarctica lipase B, sold commercially as Novozym 435, to catalyze the glycerolysis reaction, in which triglycerides exchange fatty acid groups with glycerol to form diacylglycerol. Four process variables govern the outcome: reaction time, temperature, enzyme loading, and the weight ratio of oil to glycerol. The challenge was finding the combination of these four factors that maximizes DAG content.

For decades, the standard answer to such problems has been response surface methodology, or RSM. The approach is elegant in its simplicity: run a structured set of experiments, such as a central composite design, fit a second-order polynomial equation to the results, and then use mathematics to locate the predicted optimum. In this study, the team ran a 30-experiment central composite design spanning reaction times of 9 to 13 hours, temperatures of 50 to 90 degrees Celsius, lipase loadings of 6 to 10 percent, and oil-to-glycerol ratios of 9 to 13. The fitted model was statistically strong, explaining over 96 percent of the variance, and analysis of variance confirmed that temperature, enzyme loading, and time all had significant effects.

But RSM has a fundamental weakness: it is static. Every experiment is planned before any data is collected, so if the true optimum lies outside the initial design space, or if the response surface is more twisted and nonlinear than a parabola can capture, the method has no way to adapt. Worse, when replicate measurements are averaged before modeling, RSM discards information about local variability that could reveal how trustworthy its predictions actually are. As optimization problems in food engineering grow more complex, these limitations compound.

The alternative tested here is called active learning sequential surrogate-based optimization, or AL-SSBO. Instead of committing to a fixed experimental plan, the method works in a closed loop: predict, recommend, experiment, update. An XGBoost model, a gradient-boosted ensemble of decision trees known for its computational efficiency and built-in regularization, serves as the surrogate that stands in for the real chemistry. In each of five rounds, the algorithm proposes two new experimental conditions, the lab runs them in triplicate, and the results feed back into the model before the next round begins.

The clever part lies in how the algorithm chooses what to test. Its scoring function balances exploitation, meaning conditions where the model predicts high DAG yield, against exploration, meaning conditions far from any previously sampled point, using a distance-based bonus scaled to the average spacing of the training data. A penalty discourages candidates that stray beyond a predefined distance threshold, and K-means clustering ensures the two recommendations in each round are spatially diverse rather than near-duplicates. The result is a search that leaps efficiently between distant regions of the parameter space early on, then converges tightly on the most promising zone.

The numbers tell a striking story. RSM predicted an optimum of 53.13 percent DAG at 11.78 hours, 70.96 degrees Celsius, 9 percent lipase, and an oil-to-glycerol ratio of 11.23, but experimental validation delivered only 51.70 percent, an error of 1.43 percentage points. AL-SSBO, working from the very same starting dataset, recommended conditions of 11.93 hours, 68.56 degrees Celsius, 8.93 percent lipase, and a ratio of 11.71, predicting 53.87 percent. The measured value was 53.77 percent, an error of just 0.10 percentage points, roughly fourteen times smaller than RSM’s. The yield improvement over RSM was 2.07 percentage points, achieved in a near-optimal region where the polynomial model had essentially exhausted its capability.

The team also compared surrogate models, testing Gaussian process regression and random forest alongside XGBoost under identical rules. Random forest achieved the lowest cross-validation error on the shared dataset, and Gaussian process performed best across its own sequential trajectory, but XGBoost delivered the lowest final-round prediction error and the highest experimentally achieved DAG content, so it was retained. A bootstrap-based statistical comparison, chosen because sequential data violates the independence assumptions of classical tests, showed that the 95 percent interval for the AL-SSBO optimum, spanning 52.77 to 54.16 percent, sat entirely above the RSM result, with a standardized effect size of 2.88 and zero probability of practical equivalence to the older method.

Perhaps most compelling is that the algorithm’s search path made chemical sense. Principal component analysis of the sampling trajectory revealed that early rounds explored high-temperature regions, which were quickly deprioritized once experiments showed they underperformed, while later rounds concentrated on lower temperatures, longer times, and higher enzyme and substrate ratios. This matches the underlying enzymology: moderate heat accelerates molecular collisions, but lipase is a thermally sensitive protein that degrades past a critical threshold, and longer reaction times allow the reversible glycerolysis network to approach thermodynamic equilibrium. The machine was not wandering blindly; it was rediscovering reaction kinetics from data alone, with a final-round spatial dispersion 56.6 percent lower than the initial design.

The authors are careful to note the limits of their claims. The five-round budget was fixed in advance rather than terminated by a convergence rule, retrospective checks confirmed stabilization only within the studied domain, and the results represent a laboratory-scale reference rather than an industrial operating specification. Future work must test the workflow with sparser starting datasets, during scale-up, and across multiple objectives including byproduct formation and enzyme reuse. Still, the message is clear: when a machine can learn from each experiment and decide what to try next, even a mature, well-characterized process like sunflower oil glycerolysis holds untapped potential, and the fifty-year reign of the response surface in the food lab may finally be facing a serious challenger.

Subject of Research: Machine learning-based optimization of enzymatic diacylglycerol synthesis from sunflower oil

Article Title: Active learning sequential surrogate-based optimization (AL-SSBO) versus response surface methodology in sunflower oil diacylglycerol synthesis: Performance and insight into AL-SSBO

Article References: Liu, H., Dai, Y., Liu, H., Zhang, S., Yu, H., Liao, Q., Chen, D., Jiang, X., & He, D. (2026). Active learning sequential surrogate-based optimization (AL-SSBO) versus response surface methodology in sunflower oil diacylglycerol synthesis: Performance and insight into AL-SSBO. Food Chemistry: X, 39, Article 104516. https://doi.org/10.1016/j.fochx.2026.104516

Image Credits: AI Generated

DOI: Not provided

Keywords: diacylglycerol, sunflower oil, machine learning, XGBoost, active learning, response surface methodology, enzymatic glycerolysis, lipase, food chemistry, process optimization, surrogate modeling, Novozym 435

The findings matter because lipid oxidation is one of the principal engines of food waste worldwide. When oxygen attacks the unsaturated fatty acids in oils, nuts and other lipid-containing products, it generates peroxides, aldehydes and off-flavours that render food unpalatable and, in some cases, unsafe. Food loss on this scale ripples outward, placing pressure on natural resources, raising energy consumption for distribution and storage, and driving up prices in ways that undermine food security. The global antioxidant market, valued at roughly 4.6 billion dollars in 2022 and projected to reach 7.6 billion dollars by 2032, is built largely on synthetic compounds such as BHT, BHA and TBHQ, all derived from non-renewable petrochemical feedstocks. These additives persist in the environment because of their low degradability, and their accumulating presence in processed foods has raised concerns about cumulative consumer exposure.

Essential oils offer an alternative that aligns with what the researchers describe as the e3 concept: energy efficiency, environmental sustainability and the recovery of natural additives. Unlike synthetic antioxidants, they are biodegradable, can be produced from intensively cultivated aromatic crops rather than wild harvesting, and require fewer energy inputs and no organic solvents when obtained by hydrodistillation. In the study, leaves of both species were harvested in April 2025 at the peak of flowering from the university’s experimental field in Córdoba, dried at 25 degrees Celsius, and subjected to hydrodistillation in a Clevenger-type apparatus for 60 minutes. The resulting oils were obtained in yields of 1.78 percent for Lippia turbinata and 1.96 percent for Minthostachys mollis, then analysed by gas chromatography coupled with mass spectrometry.

The chemical portraits of the two oils differ markedly. Lippia turbinata essential oil is dominated by thujone at 67.38 percent of the total composition, followed by beta-pinene at 8.52 percent and limonene at 6.51 percent, with these three compounds accounting for more than 82 percent of the oil. Minthostachys mollis oil, by contrast, is built around menthone at 51.83 percent, pulegone at 31.78 percent and isomenthone at 3.23 percent, together representing nearly 87 percent of the mixture. Neither oil is rich in the phenolic terpenes, such as carvacrol or thymol, that are conventionally associated with strong antioxidant behaviour. This makes the performance of both oils all the more intriguing, because their principal constituents act through a different and complementary antioxidant mechanism.

Phenolic antioxidants function as chain-breaking antioxidants: they donate a hydrogen atom to lipid radicals and peroxidised lipid radicals, halting the propagation of the auto-oxidation cascade. Non-phenolic terpenes such as thujone, pinene, menthone and pulegone operate as termination-enhancing antioxidants. These molecules oxidise alongside the lipids themselves, generating relatively stable radical species that pair rapidly with other radicals to form non-reactive adducts, thereby accelerating the termination reactions that remove highly reactive intermediates from the system. In the DPPH free radical scavenging assay, Minthostachys mollis oil inhibited 34.49 percent of the radical while Lippia turbinata inhibited 5.37 percent, and both oils showed total phenolic contents of roughly 8.5 to 9.1 micrograms per millilitre of gallic acid equivalents. The team cautions that such indirect tests correlate poorly with real performance in foods, partly because the polar DPPH radical interacts inefficiently with hydrophobic terpenes, which is precisely why the researchers pushed on to direct oxidation experiments in an actual food matrix.

Before testing antioxidant power, the team examined how well the oils withstand heat, since an antioxidant that decomposes during storage or processing loses its value. Aliquots of each oil were held at 60 degrees Celsius for 28 days in sealed vials, with their volatile profiles captured by solid-phase micro-extraction and analysed at five time points. Principal component analysis revealed contrasting patterns of change. The composition of Lippia turbinata oil shifted gradually and continuously across the thermal treatment: the relative abundance of thujone rose over time, not because the compound was being formed, but because less stable, higher-boiling terpenes degraded around it, while limonene and beta-pinene progressively declined as they oxidised into derivatives such as carveol and terpinen-4-ol. Minthostachys mollis oil behaved differently, with most of its compositional change occurring at the very beginning of the heat exposure. Pulegone, which contains a double bond, was progressively converted to the saturated ketone menthone, a transformation that reduced the pulegone-to-menthone ratio used by the group as an indicator of antioxidant potential. Crucially, however, neither pattern of change undermined the protective performance of the oils in the subsequent oxidation tests.

For the direct assay, the researchers fortified antioxidant-free commercial sunflower oil, which contains between 48 and 74 percent oxidation-prone linoleic acid, with 0.02 percent by weight of each essential oil, or with BHT at the equivalent of 200 parts per million, the maximum permitted in food. Samples were stored at 60 degrees Celsius for 28 days and monitored through an unusually comprehensive battery of indicators: conjugated dienes, peroxide value and anisidine value as chemical markers of primary and secondary oxidation; the composite TOTOX and INTOX values that integrate both stages; and the volatile aldehydes hexanal, (Z)-2-heptenal, (E,Z)-2,4-decadienal and (E,E)-2,4-decadienal, which arise from the fragmentation of oxidised linoleic acid and are directly responsible for rancid off-odours. By day 28, untreated oil showed dramatically higher values across nearly every indicator, while oils protected by the two essential oils recorded lower peroxide and anisidine values than even the BHT-treated samples, with Minthostachys mollis delivering the strongest protection overall.

Linear regression models fitted to the oxidation trajectories quantified the advantage with unusual precision. Using a peroxide value of 10 milliequivalents of oxygen per kilogram of oil as the regulatory shelf-life limit, unprotected sunflower oil reached the threshold in just 0.83 days. BHT extended this to 3.29 days, Lippia turbinata oil to 3.60 days, and Minthostachys mollis oil to 4.59 days, corresponding to shelf-life extensions of 3.92, 4.31 and 5.48 times the control. When shelf life was instead defined by a TOTOX value of 24, which better reflects the flavour-relevant secondary oxidation that makes oil unsellable, the essential oils again outperformed BHT, extending shelf life 3.60 and 4.23 times respectively compared with 3.36 times for the synthetic additive. Principal component analysis of all chemical and volatile indicators together explained 99.6 percent of the variability and placed the two essential-oil treatments closest together at the low-oxidation end of the spectrum, ahead of BHT.

The authors also address safety, an inevitable question for oils containing thujone and pulegone, both of which are regulated monoterpenes. Under European Union Regulation 1334/2008, pulegone is capped at 250 milligrams per kilogram in the strictest food category, and the tolerable daily intake is set at 0.1 milligrams per kilogram of body weight. At the 0.02 percent application rate used in the study, oil fortified with Minthostachys mollis contains roughly 63.56 milligrams of pulegone per kilogram, well below the European limit, and a 70-kilogram person would need to consume more than 100 grams of the oil daily to exceed the tolerable intake. Thujone from aromatic plants faces no food restrictions under the same regulation, and although a preliminary acceptable daily intake of 0.11 milligrams per kilogram has been proposed, the fortification level used here would require a person to consume nearly 50 grams of oil per day to approach that threshold, an implausible quantity for normal dietary habits.

The researchers conclude that Lippia turbinata and Minthostachys mollis essential oils demonstrate thermal stability that does not compromise their antioxidant activity, and a level of protection approaching or exceeding that of BHT under accelerated conditions. Both species are well suited to intensive cultivation in central Argentina, offering a renewable, biodegradable supply chain for natural antioxidants and a route away from extractive wild harvesting. The team emphasises that further work is needed across more complex food matrices, differing in moisture, polarity and sensory requirements, and in combination with different packaging systems, before industrial adoption can be fully realised. Still, the message is striking: two fragrant herbs from the Córdoba hills may help the food industry preserve freshness with molecules that return harmlessly to the biosphere rather than accumulating in it.

Subject of Research: Thermal stability and antioxidant performance of Lippia turbinata and Minthostachys mollis essential oils as renewable natural antioxidants in sunflower oil

Article Title: Thermal stability, and chemical and volatile oxidation indicators from essential oils of Lippia turbinata and Minthostachys mollis as potential natural antioxidants from renewable sources

Article References: Juncos, N. S., Navarro, B. D. V., Corradi, M. P., & Olmedo, R. H. (2026). Thermal stability, and chemical and volatile oxidation indicators from essential oils of Lippia turbinata and Minthostachys mollis as potential natural antioxidants from renewable sources. Discover Green Chemistry, 1(1), Article 20. https://doi.org/10.1007/s44509-026-00023-1

Image Credits: AI Generated

DOI: 10.1007/s44509-026-00023-1

Keywords: essential oils, natural antioxidants, lipid oxidation, Lippia turbinata, Minthostachys mollis, BHT, shelf life, thermal stability, sunflower oil, green chemistry, Thermal, stability