pith. machine review for the scientific record. sign in

arxiv: 2605.08365 · v1 · submitted 2026-05-08 · 💻 cs.CE

Recognition: 2 theorem links

· Lean Theorem

Texture Independently Drives Liking in AI-Generated Alternative Protein Burgers

Vahidullah Tac , Aeneas O. Koosis , Ellen Kuhl
Authors on Pith no claims yet

Pith reviewed 2026-05-12 00:47 UTC · model grok-4.3

classification 💻 cs.CE
keywords alternative proteinstexture analysissensory evaluationburger likingresiliencemeat substitutesAI-generated foodstexture profile analysis
0
0 comments X

The pith

Texture liking independently raises overall burger liking by 0.28 points, with resilience as the key mechanical driver.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines how texture affects consumer liking of alternative protein burgers made with AI assistance. By combining blind taste tests with 101 people and mechanical measurements on six different burgers, it finds that texture liking strongly predicts overall liking. Animal-based burgers stand out in sensory and mechanical profiles, while plant-based ones feel dry and crumbly, and mushroom ones springy. Resilience, a measure of how well the burger bounces back, correlates most with perceived meatiness, unlike stiffness or hardness. This suggests targeting resilience in developing sustainable meat alternatives.

Core claim

The study reveals that resilience is the strongest mechanical correlate of perceived meatiness and sensory texture in burgers, while texture liking independently predicts overall liking with a coefficient of 0.28, and meatiness dominates as a predictor of texture liking. Animal-based and hybrid burgers cluster in a firm, fatty sensory region, whereas plant-based deviate to dry and brittle, and mushroom-based to springy and gummy.

What carries the argument

Resilience from texture profile analysis, which measures the ability of the food to recover after compression, serving as the link between mechanical properties and sensory perception of meatiness.

If this is right

  • Improving resilience in alternative proteins could enhance perceived meatiness and texture liking.
  • Texture should be a primary design goal alongside flavor for sustainable burgers.
  • Hybrid animal-mushroom patties can achieve sensory profiles similar to animal-based ones.
  • AI-generated burgers can be optimized for texture to match consumer preferences.
  • Stiffness and hardness are not useful targets for texture engineering in this context.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Targeting resilience might help reduce the gap between alternative and conventional meat products in consumer preference.
  • The results could inform processing methods to adjust resilience in plant-based ingredients.
  • Extending these findings to other food formats such as sausages or ground meats may reveal broader applications for texture design.
  • If validated, this approach could contribute to more sustainable food systems by making plant-based options more appealing.

Load-bearing premise

Mechanical measurements of texture like resilience directly reflect what people sense and prefer in blind tastings, without hidden influences from flavor or how the test is run.

What would settle it

A follow-up test in which burgers are adjusted only in resilience while other properties stay fixed, yet meatiness ratings and overall liking scores show no corresponding shift among blind tasters.

Figures

Figures reproduced from arXiv: 2605.08365 by Aeneas O. Koosis, Ellen Kuhl, Vahidullah Tac.

Figure 1
Figure 1. Figure 1: Overview of the six burgers used in this study. We generated five new burgers using our generative artificial intelligence platform for burgers: delicious burger 1 and delicious burger 2 were optimized for palatability and contain animal meat; sustainable burger 1 was optimized for sustainability and uses mushroom; sustainable burger 2 uses a hybrid animal-mushroom blend; and nutritious burger uses beans. … view at source ↗
Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Mean sensory ratings for overall liking, flavor, texture, and five texture-related at￾tributes. The plant-based burgers, sustainable burger 1 and nutritious burger, score lower on overall liking, flavor, texture, hardness, fattiness, and meatiness, but score higher on softness and fibrousness, compared to the animal-based burgers. Sensory evaluation across 101 participants reveals clear differences between… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of texture profile analysis parameters and sensory texture ratings. Me￾chanical properties span wide ranges across all six burgers, while sensory ratings remain compressed. The order of burgers along the mechanical parameters does not always mirror the order along sensory ratings. from this animal-based cluster in distinctly different directions: the plant-based nutritious burger scores low on P… view at source ↗
Figure 5
Figure 5. Figure 5: Correspondence analysis biplot that relates burgers, sensory texture attributes, and texture profile analysis parameters in a shared low-dimensional space. Meat burgers form a cluster in the upper right quadrant along with sensory attributes such as firm, holds together, and fatty, and most texture profile analysis parameters. The plant-based nutritious burger has a lower value on principal component 1 (PC… view at source ↗
Figure 6
Figure 6. Figure 6: Linear correlations between texture profile analysis parameters and sensory attributes. The data suggest that resilience is particularly promising for texture engineering, as it has significant cor￾relations with four sensory attributes; crispy/crunchy, holds together, mushy and tough. Cohesiveness, springiness and chewiness also have significant correlations with sensory attributes. However stiffness and … view at source ↗
Figure 7
Figure 7. Figure 7: Correlation between texture liking and textural attributes. Texture liking is rated on a 7-point Likert scale, textural attributes are rated on a 5-point Likert scale. Meatiness shows the strongest association with perceived texture quality with a 1-point increase in meatiness independently explaining 0.59 point increase in texture liking in linear mixed-effects modeling. Among the six texture profile anal… view at source ↗
Figure 8
Figure 8. Figure 8: Relationship between texture profile analysis parameters and perceived meatiness. Resilience shows a strong positive correlation with meatiness, while stiffness, hardness, cohesiveness, springi￾ness, and chewiness show no statistically significant association. Funding This research was supported by the Schmidt Science Fellowship in partnership with the Rhodes Trust to Vahidullah Tac, and by the Stanford Bi… view at source ↗
Figure 9
Figure 9. Figure 9: Mixed-effects model shows the relationship between overall liking, flavor, and texture. Texture has a strong impact in determining overall liking. For the same flavor score, a 1-point increase in texture score results in a 0.28-point increase in overall score, on average. [3] J. Poore and T. Nemecek. Reducing food’s environmental impacts through producers and consumers. Science, 360(6392):987–992, 2018. [4… view at source ↗
read the original abstract

Texture shapes how we perceive and like food, yet clear links between mechanical measurements and sensory perception of texture remain elusive. Here we combine sensory data from a blind tasting with 101 participants with mechanical texture profile analysis across six burgers to identify the textural features that drive consumer perception and liking. We compare five burgers -- generated with artificial intelligence -- with animal-based, plant-based, mushroom-based, and hybrid animal-mushroom patties, and the classical Big\,Mac. Three main findings emerge: First, animal-based burgers occupy a distinctive and coherent sensory-mechanical region associated with attributes such as firm, fatty, and holds together. Second, mushroom- and plant-based burgers deviate from this region in protein-dependent ways: mushroom-based burgers associate with springy and gummy textures, while plant-based burgers associate with dry, brittle, and crumbly textures. Hybrid animal-mushroom burgers, however, maintain sensory profiles comparable to fully animal-based burgers. Third, resilience emerges as the strongest mechanical correlate of perceived meatiness and sensory texture, while stiffness and hardness show no statistically significant association with consumer perception. Texture independently predicts overall liking alongside flavor: increasing texture liking by one point increases overall liking by 0.28. Among all sensory attributes, meatiness is the dominant predictor of texture liking. These findings identify resilience as a promising target for texture engineering and establish texture as a critical design objective for sustainable alternative proteins.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript reports an empirical study combining blind sensory evaluations from 101 participants across six burger products (AI-generated alternatives, animal-based, plant-based, mushroom-based, hybrid, and Big Mac) with mechanical texture profile analysis (TPA). It claims three main results: (1) animal-based burgers occupy a distinct sensory-mechanical region (firm, fatty, holds together) while mushroom-based are springy/gummy and plant-based dry/brittle/crumbly, with hybrids comparable to animal-based; (2) resilience is the strongest mechanical correlate of perceived meatiness and sensory texture, whereas stiffness and hardness show no statistically significant association; (3) texture liking independently predicts overall liking (increasing texture liking by one point raises overall liking by 0.28), with meatiness as the dominant predictor of texture liking. These support resilience as a target for texture engineering in alternative proteins.

Significance. If the associations hold after addressing sample-size and methodological limitations, the work provides actionable, evidence-based guidance for engineering texture in sustainable alternative proteins, particularly by prioritizing resilience over other TPA parameters and treating texture as separable from flavor in driving liking. The combination of human sensory data with objective mechanical measurements and the inclusion of AI-generated samples are strengths that could inform both product development and future modeling in food science.

major comments (3)
  1. [Results (mechanical-sensory correlations)] Results section on mechanical-sensory correlations: TPA parameters are measured once per burger type, yielding only n=6 independent observations when correlating with participant-averaged sensory scores. This small sample directly undermines the reliability of designating resilience as the 'strongest' correlate and asserting null associations for stiffness and hardness, as both the ranking and the lack of significance are sensitive to the specific choice of the six burgers, measurement noise, and low statistical power.
  2. [Methods and Results (statistical analysis)] Methods and Results on regression and statistical reporting: The reported 0.28 coefficient for texture liking predicting overall liking, along with other associations, lacks specification of the exact model (e.g., linear regression vs. mixed-effects accounting for participant variability), error bars or confidence intervals, data exclusion criteria, and any correction for multiple comparisons across attributes and parameters. These omissions prevent full assessment of the robustness of the central claims.
  3. [Discussion] Discussion of mechanical-to-sensory mapping: The interpretation that TPA resilience directly drives perceived meatiness and texture assumes minimal unmeasured confounders (e.g., flavor-texture interactions or residual biases in the blind tasting protocol). Given that mechanical measurements are single per type and sensory data are aggregated, additional justification or sensitivity analyses are needed to support the texture-engineering recommendations.
minor comments (2)
  1. [Abstract] The abstract contains a LaTeX artifact ('Big,Mac'); correct to 'Big Mac' for clarity.
  2. [Figures/Tables] Figures or tables presenting correlations should include individual data points (given n=6) and error bars to allow readers to evaluate the strength and variability of the reported associations.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments, which have helped us strengthen the manuscript. We address each major point below and have revised the text to improve clarity, statistical reporting, and discussion of limitations.

read point-by-point responses

  1. Referee: Results section on mechanical-sensory correlations: TPA parameters are measured once per burger type, yielding only n=6 independent observations when correlating with participant-averaged sensory scores. This small sample directly undermines the reliability of designating resilience as the 'strongest' correlate and asserting null associations for stiffness and hardness, as both the ranking and the lack of significance are sensitive to the specific choice of the six burgers, measurement noise, and low statistical power.

    Authors: We agree that the mechanical-sensory correlations rely on n=6 observations, which inherently limits statistical power and makes rankings and null findings sensitive to sample composition. The sensory scores are robustly averaged across 101 participants, but this does not increase the number of independent mechanical data points. In the revision we have added an explicit discussion of this limitation in the Results and Discussion sections, qualified the claims about null associations for stiffness and hardness as exploratory, and included leave-one-out sensitivity checks to demonstrate that the resilience correlation remains the strongest across subsets. We retain the original ranking as descriptive of the observed data while acknowledging the small-n constraint. revision: yes

  2. Referee: Methods and Results on regression and statistical reporting: The reported 0.28 coefficient for texture liking predicting overall liking, along with other associations, lacks specification of the exact model (e.g., linear regression vs. mixed-effects accounting for participant variability), error bars or confidence intervals, data exclusion criteria, and any correction for multiple comparisons across attributes and parameters. These omissions prevent full assessment of the robustness of the central claims.

    Authors: We accept that the statistical reporting was insufficiently detailed. The 0.28 coefficient derives from ordinary least-squares linear regression on participant-averaged scores (not mixed-effects models, as participant-level variability was already summarized). We have revised the Methods section to specify the models, added 95% confidence intervals for all reported coefficients, stated that no observations were excluded beyond standard data-quality checks, and noted that analyses were exploratory without formal multiple-comparison correction. These changes allow readers to evaluate robustness directly. revision: yes

  3. Referee: Discussion of mechanical-to-sensory mapping: The interpretation that TPA resilience directly drives perceived meatiness and texture assumes minimal unmeasured confounders (e.g., flavor-texture interactions or residual biases in the blind tasting protocol). Given that mechanical measurements are single per type and sensory data are aggregated, additional justification or sensitivity analyses are needed to support the texture-engineering recommendations.

    Authors: We recognize that single mechanical measurements per type and aggregated sensory data leave room for unmeasured confounders such as flavor-texture interactions. The blind protocol reduced visual and labeling biases, but residual effects cannot be ruled out. In the revised Discussion we have added a paragraph on these assumptions, included partial-correlation sensitivity checks controlling for other sensory attributes, and tempered the engineering recommendations to emphasize that resilience is a promising target pending confirmation in larger, multi-measurement studies. This provides the requested justification while preserving the core interpretation supported by the data. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical correlations from independent sensory and mechanical data

full rationale

The paper reports statistical associations from blind sensory tasting (n=101 participants) and mechanical TPA measurements on six distinct burger formulations. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains appear in the abstract or described methods. Claims such as resilience as the strongest correlate or texture independently predicting liking are direct outputs of the external data collection rather than reductions to the paper's own inputs by construction. The analysis is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard statistical assumptions for sensory-mechanical correlations and regression; no free parameters, ad-hoc axioms, or invented entities are introduced beyond conventional food science measurement protocols.

axioms (2)
  • domain assumption Mechanical texture profile analysis measurements validly capture sensory texture attributes
    Invoked when linking lab compression tests to participant ratings of firmness, springiness, and meatiness.
  • domain assumption Blind tasting with 101 participants yields unbiased liking and attribute scores
    Assumed when interpreting overall liking predictions and meatiness dominance.

pith-pipeline@v0.9.0 · 5553 in / 1371 out tokens · 45214 ms · 2026-05-12T00:47:27.209941+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages · 1 internal anchor

  1. [1]

    Van Den Berg

    Jovan Ili´ c, Ilija Djekic, Igor Tomasevic, Filip Oosterlinck, and Marco A. Van Den Berg. Materials Properties, Oral Processing, and Sensory Analysis of Eating Meat and Meat Analogs.Annual Review of Food Science and Technology, 13(1):193–215, March 2022

    work page 2022

  2. [2]

    Clark, M

    M. Clark, M. Springmann, M. Rayner, P. Scarborough, J. Hill, D. Tilman, J. I. Macdiarmid, J. Fanzo, L. Bandy, and R. A. Harrington. Estimating the environmental impacts of 57,000 food products. Proceedings of the National Academy of Sciences, 119(33):e2120584119, 2022. 11 1 7flavor score 1 7overall score = 0.65 1 7texture score 1 7overall score = 0.28 1 7...

    work page 2022

  3. [3]

    Poore and T

    J. Poore and T. Nemecek. Reducing food’s environmental impacts through producers and consumers. Science, 360(6392):987–992, 2018

    work page 2018

  4. [4]

    Willett, J

    W. Willett, J. Rockstr¨ om, B. Loken, et al. Food in the anthropocene: the eat–lancet commission on healthy diets from sustainable food systems.The Lancet, 393(10170):447–492, 2019

    work page 2019

  5. [5]

    Malila, I.O

    Y. Malila, I.O. Owolabi, T. Chotanaphuti, et al. Current challenges of alternative proteins as future foods.npj Science of Food, 8:1–16, 2024

    work page 2024

  6. [6]

    S. R. St. Pierre and E. Kuhl. Mimicking mechanics: A comparison of meat and meat analogs.Foods, 13(21):3495, 2024

    work page 2024

  7. [7]

    Generative Artificial Intelligence creates deli- cious, sustainable, and nutritious burgers, February 2026

    Vahidullah Tac, Christopher Gardner, and Ellen Kuhl. Generative Artificial Intelligence creates deli- cious, sustainable, and nutritious burgers, February 2026

    work page 2026

  8. [8]

    Generative AI for material design: A mechanics perspective from burgers to matter, April 2026

    Vahidullah Tac and Ellen Kuhl. Generative AI for material design: A mechanics perspective from burgers to matter, April 2026

    work page 2026

  9. [9]

    A. S. Szczesniak. Texture is a sensory property.Food Quality and Preference, 13(4):215–225, 2002

    work page 2002

  10. [10]

    van den Bedem, E

    S. van den Bedem, E. Kuhl, and C. Cotto. Open-source benchmarking of plant-based and animal meats. 2026

    work page 2026

  11. [11]

    J. Chen. Food oral processing – a review.Food Hydrocolloids, 23(1):1–25, 2009

    work page 2009

  12. [12]

    R. A. Dunne, E. C. Darwin, V. A. Perez Medina, et al. Texture profile analysis and rheology of plant-based and animal meats.Food Research International, 198:115234, 2025

    work page 2025

  13. [13]

    P. P. Purslow. The structure and functional significance of variations in the connective tissue within muscle.Comparative Biochemistry and Physiology Part A, 133:947–966, 2002. 12

    work page 2002

  14. [14]

    B. L. Dekkers, R. M. Boom, and A. J. van der Goot. Structuring processes for meat analogues.Trends in Food Science & Technology, 81:25–36, 2018

    work page 2018

  15. [15]

    Thibault Vervenne, Skyler R. St. Pierre, Nele Famaey, and Ellen Kuhl. Probing mycelium mechanics and taste: The moist and fibrous signature of fungi steak.Acta Biomaterialia, 202:341–351, August 2025

    work page 2025

  16. [16]

    Bahador Bahmani and WaiChing Sun. A kd-tree-accelerated hybrid data-driven/model-based approach for poroelasticity problems with multi-fidelity multi-physics data.Computer Methods in Applied Me- chanics and Engineering, 382:113868, August 2021

    work page 2021

  17. [17]

    Visco-hyperelastic constitutive modeling of strain rate sensitive soft materials.Journal of the Mechanics and Physics of Solids, 135:103777, February 2020

    Kshitiz Upadhyay, Ghatu Subhash, and Douglas Spearot. Visco-hyperelastic constitutive modeling of strain rate sensitive soft materials.Journal of the Mechanics and Physics of Solids, 135:103777, February 2020

    work page 2020

  18. [18]

    William M. Breene. Application of texture profile analysis to instrumental food texture evaluation. Journal of Texture Studies, 6(1):53–82, 1975

    work page 1975

  19. [19]

    Malcolm C. Bourne. Texture profile analysis.Food Technology, 32(7):62–66, 1978

    work page 1978

  20. [20]

    S Novakovi´ c and I Tomaˇ sevi´ c. A comparison between Warner-Bratzler shear force measurement and texture profile analysis of meat and meat products: A review.IOP Conference Series: Earth and Environmental Science, 85:012063, September 2017

    work page 2017

  21. [21]

    Skyler R. St. Pierre and Ellen Kuhl. Mimicking mechanics: A comparison of meat and meat analogs. Foods, 13(21):3495, October 2024

    work page 2024

  22. [22]

    Qian, S.I

    C. Qian, S.I. Murphy, R.H. Orsi, and M. Wiedmann. How Can AI Help Improve Food Safety?Annual Review of Food Science and Technology, 14(1):517–538, March 2023

    work page 2023

  23. [23]

    A Physics-Augmented Machine Learning Constitutive Model for Damage in Solids, July 2025

    Amirhossein Amiri-Hezaveh and Adrian Buganza Tepole. A Physics-Augmented Machine Learning Constitutive Model for Damage in Solids, July 2025

    work page 2025

  24. [24]

    H. M. Nazari, C. E. Ozdemir, and M. Tyagi. Reduced-order modeling of temporal local scour dynamics beneath a submerged cylinder in a steady flow using autoencoders.Physics of Fluids, 37(8):083389, August 2025

    work page 2025

  25. [25]

    Thomas, Eduardo Barocio, and R

    Akshay J. Thomas, Eduardo Barocio, and R. Byron Pipes. A machine learning approach to deter- mine the elastic properties of printed fiber-reinforced polymers.Composites Science and Technology, 220:109293, March 2022

    work page 2022

  26. [26]

    Argmax flows and multinomial diffusion: Learning categorical distributions

    Emiel Hoogeboom, Didrik Nielsen, Priyank Jaini, Patrick Forr´ e, and Max Welling. Argmax flows and multinomial diffusion: Learning categorical distributions. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, volume 34, pages 12454–12465, 2021

    work page 2021

  27. [27]

    Rausch, Ilias Bilionis, Francisco Sahli Costabal, and Adrian Buganza Tepole

    Vahidullah Ta¸ c, Manuel K. Rausch, Ilias Bilionis, Francisco Sahli Costabal, and Adrian Buganza Tepole. Generative hyperelasticity with physics-informed probabilistic diffusion fields.Engineering with Com- puters, 41:51–69, May 2024

    work page 2024

  28. [28]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations.arXiv, page doi:10.48550/arXiv.2011.13456, 2021

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2011.13456 2011

  29. [29]

    Pierre, EC Darwin, D Adil, MC Aviles, A Date, RA Dunne, Y Lall, M Parra Vallecillo, VA Perez Medina, K Linka, ME Levenston, and E Kuhl

    SR St. Pierre, EC Darwin, D Adil, MC Aviles, A Date, RA Dunne, Y Lall, M Parra Vallecillo, VA Perez Medina, K Linka, ME Levenston, and E Kuhl. The mechanical and sensory signature of plant-based and animal meat.npj Science of Food, 8:94, 2024

    work page 2024

  30. [30]

    Likert-Type Scale.Encyclopedia, 5(1):18, February 2025

    Malcolm Koo and Shih-Wei Yang. Likert-Type Scale.Encyclopedia, 5(1):18, February 2025. 13

    work page 2025

  31. [31]

    Correspondence Analysis

    Herv´ e Abdi and Michel B´ era. Correspondence Analysis. In Reda Alhajj and Jon Rokne, editors, Encyclopedia of Social Network Analysis and Mining, pages 1–12. Springer New York, New York, NY, 2017

    work page 2017

  32. [32]

    Max Halford. Prince. Python package, MIT License, accessed April 2026

    work page 2026

  33. [33]

    Analytical Methods for Social Research

    Andrew Gelman and Jennifer Hill.Data Analysis Using Regression and Multilevel/Hierarchical Models. Analytical Methods for Social Research. Cambridge University Press, Cambridge, 2006

    work page 2006

  34. [34]

    statsmodels: Econometric and statistical modeling with python

    Skipper Seabold and Josef Perktold. statsmodels: Econometric and statistical modeling with python. In9th Python in Science Conference, 2010. 14

    work page 2010