The research tackles a long-standing bottleneck. Food companies developing plant-based products typically rely on trial-and-error experimentation, because the relationship between formulation and final texture is nonlinear and context-dependent. Machine learning has recently been recruited to help. Earlier work demonstrated that hardness and chewiness of plant-based meat analogues can be estimated from proximate composition using supervised learning, and that machine-learning-assisted optimization can reduce experimental burden in high-moisture extrusion. But these models share a critical weakness: they are sensitive to dataset size, feature selection, and something statisticians call distributional shift. In food research, where experiments are expensive and sample numbers are small, new formulations often fall outside the range of the training data, and purely statistical models tend to fail exactly when researchers need them most.

Ma’s study compared three fundamentally different prediction approaches across two held-out prediction tasks. The first approach used standard machine learning baselines: gradient boosting, random forest, and TabPFN, a pre-trained transformer model suited to small tabular datasets. The second was pure inverse distance weighting, or IDW, a retrieval method that predicts each new sample by averaging the three most similar training samples in normalized composition space. The third was a knowledge-based AI agent that used the same retrieval logic but added an explicit layer of food-domain reasoning, drawing on a curated mechanistic knowledge base of literature-derived rules. The comparison was deliberately fair: all three methods received identical input features, so any difference in performance could be attributed to the mechanistic knowledge the agent carried with it.

The first case study used a dataset of 54 plant-based meat analogue samples characterized by seven compositional features, including protein, fat, carbohydrate, fibre, ash, moisture, and target moisture, alongside measured hardness and chewiness. To create a stringent test, the researcher held out 13 samples whose fibre and protein levels deliberately fell outside the training range, leaving only 41 samples for training. This out-of-distribution design mimics a common real-world scenario: a food developer introduces a novel protein source or an unusual fibre level, and existing models must extrapolate rather than interpolate. The prediction targets were hardness, the first compression peak force in texture profile analysis, and chewiness, the mechanical energy required to chew a sample to a swallowable state.

The results revealed a dramatic hierarchy. The knowledge-based agent achieved the lowest error for hardness, with a mean absolute error of 4.66 newtons and a coefficient of determination of 0.782, and for chewiness it reached a mean absolute error of 4.45 joules. Gradient boosting, the best machine learning baseline, was statistically comparable, though the agent was numerically ahead. Pure IDW retrieval, by contrast, performed catastrophically, with strongly negative R-squared values meaning its predictions were less informative than simply guessing the training average. The paired differences between the agent and pure retrieval were large and statistically significant, roughly 14 newtons for hardness and 11 joules for chewiness, with confidence intervals comfortably excluding zero. Notably, some machine learning models were highly unstable on the small, shifted test set, and gradient boosting behaved partly like a classifier, grouping its hardness predictions into three discrete levels that failed to capture the variation in the new formulations.

What made the agent’s success possible was the mechanistic knowledge base built into its reasoning stage. The agent, powered by a large language model, operated in two stages. First, it retrieved the three nearest training analogues using the same normalized Euclidean distance as the IDW baseline. Second, it processed these retrieved examples together with compact, literature-derived rules for plant-based meat analogues, including moisture-dependent protein network formation, the relative texturizing capacity of pea, soy, and wheat proteins under high-moisture extrusion, fibre-reinforced network stiffening, and fat-induced lubrication and softening. The model then chose an adjustment within a qualitative range defined by these rules, and when no rule applied, it simply emitted the retrieval-weighted average. Crucially, the test set remained entirely inaccessible during knowledge compilation, and no measured texture values of test samples were included in the retrieval base, ensuring strict train-test separation.

The second case study tested whether the approach would hold up on a larger dataset with different physics. It used 311 measurements on plant protein-polysaccharide mixtures, split into 246 training and 63 test samples, with two structural outliers excluded on data-inspection grounds before any predictions were run, after their maximum-stress values of roughly 4 to 7 pascals proved about three orders of magnitude below the rest of the distribution, indicating measurement failure. The prediction targets were the storage modulus, a small-deformation measure of gel stiffness, and the maximum stress at 200 percent strain, a large-deformation failure property. Here, all methods performed better thanks to the larger sample size, but the agent still came out on top. For the storage modulus it achieved a mean absolute error of 152.30 kilopascals and an R-squared of 0.695, edging out the best machine learning model, TabPFN. The gap widened dramatically for maximum stress, where machine learning models struggled badly, with the best one reaching only an R-squared of 0.412 and predictions concentrated in a narrow band that missed the full dynamic range of the test set. The agent posted a mean absolute error of 17.21 kilopascals and an R-squared of 0.765, following the parity line most closely across the measurable range.

Ablation tests confirmed where the agent’s advantage came from. When the mechanistic knowledge base was removed, leaving the language model with retrieval alone, its performance collapsed back to the IDW baseline in the meat analogue case and even fell below IDW for maximum stress. In other words, the improvement stemmed from the encoded food-science knowledge, not from the language model’s general reasoning abilities. This finding aligns with broader trends in artificial intelligence research, where retrieval-augmented and tool-using systems increasingly combine data-driven lookup with explicit reasoning, and it suggests a practical recipe: gray-box prediction that marries empirical data with mechanistic understanding, rather than relying on either alone.

The implications for food formulation are considerable. Predicting mechanical behavior from composition alone has always been fragile because texture depends on how water and biopolymer networks reorganize during processing, information that numeric feature vectors rarely capture. Knowledge-guided retrieval offers a way to stabilize predictions precisely under distribution shift, the condition that breaks conventional models, and could accelerate AI-assisted formulation screening for plant-based food design. The study also acknowledges limits: only two tabular case studies with scalar targets were examined, the agent’s quality depends on the accuracy of its encoded knowledge, and the implementation is not self-updating, so new observations must be added manually. Future work, Ma suggests, should explore continuous learning frameworks in which the agent autonomously refreshes its knowledge base as new experiments arrive, extending knowledge-guided retrieval to broader food systems, processing conditions, and mechanical targets.

Subject of Research: Comparing knowledge-based AI agents and machine learning for predicting mechanical and rheological properties of dense food structures

Article Title: Comparing AI agents and machine learning for predicting mechanical and rheological properties of dense food structures

Article References: Ma, Y. (2026). Comparing AI agents and machine learning for predicting mechanical and rheological properties of dense food structures. Current Research in Food Science, 13, Article 101578. https://doi.org/10.1016/j.crfs.2026.101578

Image Credits: AI Generated

DOI: 10.1016/j.crfs.2026.101578

Keywords: artificial intelligence, machine learning, plant-based meat analogues, food texture, rheology, food science, retrieval-augmented prediction, protein-polysaccharide gels, high-moisture extrusion, food formulation, large language models, mechanistic knowledge

Moshe H. Azachi and Zeev Wiesman compared two commercially available products that could hardly be more different in structure: an Angus beef burger and a soy-based vegan burger. Both were cooked in an air fryer to an internal temperature of 72 degrees Celsius, then probed with an arsenal of techniques including low-field time-domain nuclear magnetic resonance (TD-NMR), cryogenic scanning electron microscopy, texture profile analysis, and a trained sensory panel. The question was deceptively simple: could measurements of the water inside the burger, taken before anyone chewed it, explain which one would taste juicier?

The answer hinged on a subtle but crucial distinction in physics. When a burger is chewed, the teeth compress and fracture a hydrated soft material, generating pressure gradients that squeeze fluid through interconnected pathways. This mechanically driven serum release is what food scientists increasingly regard as the true proximal driver of perceived juiciness. Molecular self-diffusion, by contrast, is the spontaneous, thermally driven jiggling and translational motion of individual water molecules, measured by pulsed-field-gradient NMR in a completely undeformed sample. The two processes are physically distinct, but the researchers hypothesized they might share common structural determinants: the size of water-filled domains, the connectivity of the aqueous network, tortuosity, interfaces, and the strength of water-protein interactions.

The first surprise came from the moisture numbers. The beef burgers started with 68.05 percent water and the soy burgers with 62.61 percent, a difference that did not reach statistical significance. During cooking, the beef patties lost far more mass, shedding 35.73 percent of their weight compared with 21.21 percent for the soy burgers, and shrank nearly twice as much in diameter. By conventional food-science logic, the soy burger should have been the juicier product: it held onto its water more tenaciously. Instead, the opposite happened. A panel of thirteen trained assessors, scoring five juiciness-related attributes on a ten-point scale, rated the beef burger dramatically higher on every single one, from first-bite juiciness to sustained juiciness, rate of juice release, and mouthcoating. The composite Sensory Juiciness Index came to 7.46 for beef versus 5.00 for the soy burger, a difference the authors report as highly significant.

The decisive instrumental signal came from the NMR diffusion measurements. Using a 20 MHz Bruker Minispec mq20 and the classic Stejskal-Tanner pulsed-field-gradient sequence, the team measured an apparent water self-diffusion coefficient in both the interior and the cooked outer layer of each burger. The beef interior showed the highest value, 1.03 by ten to the minus nine square meters per second, while the soy burger’s outer region showed the lowest, just 0.36 by ten to the minus nine. At the whole-burger level, beef averaged 1.030 and soy 0.704 by ten to the minus nine square meters per second, a difference with a p-value below 0.001. In both products, diffusion decreased from the moist interior toward the dehydrated surface, but the formulation gap dwarfed the regional one.

What makes this result conceptually important is what did not differentiate the burgers. Transverse relaxation times, the T2 values that reflect local molecular environments and confinement, told a murkier story. The longest relaxation components exceeded roughly 300 milliseconds in both matrices and did not differ significantly, and the mean long-T2 values at the parent-burger level were statistically indistinguishable. The soy burger’s outer layer was especially instructive: it combined a comparatively long mono-exponential T2 with a very low diffusion coefficient, demonstrating that water molecules can be locally mobile yet still encounter severe barriers to longer-range displacement. Relaxation and diffusion, in other words, probe different aspects of the water state and cannot be used interchangeably.

Cryogenic scanning electron microscopy provided visual context. The researchers flash-froze small sections from the interior and outer regions of each cooked burger, fractured them under cryogenic conditions, etched them, and imaged them with an in-lens detector at around minus 120 degrees Celsius. The beef samples revealed a heterogeneous architecture of irregular pore-like and hydrated domains with apparently greater continuity, while the soy matrix appeared denser and more homogeneous, with thicker solid regions and lower apparent connectivity, especially in the cooked outer layer. This qualitative morphology paralleled the spatial ordering of the diffusion coefficients, though the authors are careful to note that pore size, mesh dimensions, tortuosity, and three-dimensional connectivity were not quantified, so the structural explanation remains a hypothesis rather than a demonstrated mechanism.

Texture told a similarly incomplete story. The beef burgers were numerically harder and chewier than the soy burgers, though these differences did not reach significance with only three independent parent burgers per formulation, and only cohesiveness differed significantly. The juiciness gap therefore cannot be reduced to a simple soft-versus-hard distinction. What the study does support is a multiscale framework: food architecture shapes the pre-deformation molecular state of water, which TD-NMR can measure non-destructively; during mastication, deformation generates pressure gradients that drive macroscopic serum displacement through whatever pathways the architecture permits; and that released serum lubricates the mouth, producing the sensation of juiciness. The intermediate links in this chain, the actual serum flux and oral lubrication, were not measured here and remain testable propositions for future work.

The authors are equally candid about the limits of their evidence. Because the sensory ratings and the NMR measurements were obtained from separate experimental units rather than matched observations from the same parent burgers, no within-sample correlation between diffusion and juiciness could be calculated. The findings represent concordant formulation-level contrasts, not proof of causation. The study also compared only one commercial beef product and one commercial soy product, so it establishes a contrast between two matrices rather than a universal law of animal versus plant. An exploratory Water Accessibility Index, defined as the ratio of diffusion to relaxation, showed a significant difference between formulations but remains an empirical, unit-dependent descriptor pending external validation.

Even so, the implications for food designers are concrete. Strategies aimed solely at maximizing water-holding capacity or minimizing cooking loss may be fundamentally insufficient: a highly hydrated matrix can still taste dry if its architecture strongly restricts the redistribution of fluid during chewing. The design objective, the authors argue, is not simply to trap water but to retain it through manufacture and cooking while making an appropriate fraction available at the moment of mechanical deformation. Variables such as protein-network organization, domain connectivity, structural anisotropy, lipid distribution, and protein-polysaccharide interactions all become levers worth pulling. And because TD-NMR diffusion is rapid and non-destructive, it could serve as a screening tool that predicts, before a single taste test, whether a reformulated plant-based patty has the architectural prerequisites for a genuinely juicy bite.

Subject of Research: Water self-diffusion measured by TD-NMR in cooked beef and soy-based burger matrices and its relationship to sensory juiciness

Article Title: Water self-diffusion differentiates two protein-based burger matrices with contrasting sensory juiciness

Article References: Azachi, M. H., & Wiesman, Z. (2026). Water self-diffusion differentiates two protein-based burger matrices with contrasting sensory juiciness. Current Research in Food Science, 13, Article 101562. https://doi.org/10.1016/j.crfs.2026.101562

Image Credits: AI Generated

DOI: 10.1016/j.crfs.2026.101562

Keywords: water self-diffusion, TD-NMR, plant-based meat analogues, juiciness, food microstructure, soy protein, beef burgers, sensory evaluation, serum release, Cryo-SEM, food oral processing, water-holding capacity