Systems and methods for producing fermented comestibles with precisely controlled ethanol and metabolite profiles with machine learning models

Abstract

A computer-implemented method is provided for producing a fermented beverage having an alcohol content below a predefined threshold includes receiving a target organoleptic profile corresponding to desired flavor and aroma characteristics. The method applies a trained machine-learning model to evaluate candidate combinations including at least one precursor substrate formulation, at least one food-safe microbial strain or consortium, and associated fermentation process parameters. The trained machine-learning model predicts, for each candidate combination, an organoleptic profile approximating the target profile and a final alcohol concentration below the predefined threshold. The method selects a precursor substrate formulation, microbial strain or consortium, and fermentation process parameters based on the model's output. The selected precursor substrate formulation is fermented with the selected microbial strain or consortium in accordance with the selected fermentation process parameters to produce the fermented beverage without post-fermentation alcohol removal.

Description

SYSTEMS AND METHODS FOR PRODUCING FERMENTED COMESTIBLES WITH PRECISELY CONTROLLED ETHANOL AND METABOLITE PROFILES WITH MACHINE LEARNING MODELSCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Patent Application No.63 / 740,959 titled SYSTEM AND METHOD FOR OPTIMIZING NON-ALCOHOLIC BEVERAGE FERMENTATION PROCESSES BASED ON ARTIFICIAL INTELLIGENCE MACHINE LEARNING MODELS, filed December 31, 2024, which is hereby incorporated by reference in its entirety. This application is related to PCT Application No. PCT / IB2025 / 059740, titled SYSTEM AND METHOD FOR PRESCRIBING FERMENTATION BASED ON MACHINE LEARNING MODELS, filed September 27, 2025, which is hereby incorporated by reference in its entirety.FIELD OF INVENTION

[0002] The present disclosure relates to fermented comestibles and methods for their production, and more particularly to computer-implemented methods employing trained machine-learning models to produce non-alcoholic or zero-alcohol fermented comestibles.BACKGROUND

[0003] The market for sophisticated non-alcoholic or zero-alcohol fermented comestibles is growing. Specifically, the market for beverages of the same is growing rapidly as consumers seek complex, adult-oriented alternatives to traditional alcoholic beverages. Consumers increasingly demand products that approximate the depth, balance, and sensory complexity of beer, wine, cider, and spirits, without the intoxicating effects of alcohol. At the same time, there is growing emphasis on low sugar content, clean labels, and extended shelf life. Existing nonalcoholic offerings frequently fall short of these expectations, particularly with respect to flavor consistency and complexity. Beverages such as kombucha often exhibit variable alcohol content, inconsistent sensory profiles, and high residual sugar. These challenges make such products difficult to control, reproduce, and scale commercially.

[0004] Fermented comestibles, including beverages such as beer, wine, cider, and spirits, and food products such as yogurt and cheese, are traditionally produced through carefully controlled fermentation processes. These processes begin with the selection of fermentationagents, primarily yeasts and, in some cases, bacteria, which convert sugars into ethanol and other metabolic byproducts. The choice of fermentation agent directly influences the flavor, aroma, mouthfeel, and overall quality of the final product. Process parameters such as temperature, pH, oxygen availability, and nutrient composition are meticulously controlled to achieve consistent results. In conventional alcoholic fermentation, ethanol production is closely coupled with the formation of desirable flavor-active metabolites. This coupling presents a fundamental challenge when attempting to produce beverages with low or negligible alcohol content.

[0005] Known methods for producing non-alcoholic or alcohol-free beverages often require post-fermentation alcohol removal, such as vacuum distillation or reverse osmosis. In some instances, these methods can strip volatile aroma compounds, alter mouthfeel, and increase production costs and energy consumption. In addition, shelf-stability can be an issue. Therefore, a need exists to address some of these or other disadvantages.SUMMARY

[0006] This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

[0007] According to an aspect of the present disclosure, a computer-implemented method for producing a fermented comestible having an alcohol content below a predefined threshold is provided. The method includes receiving a target organoleptic profile corresponding to desired flavor and aroma characteristics. The method further includes applying a trained machine-learning model to evaluate candidate fermentation configurations by generating, for each candidate fermentation configuration, an organoleptic profile and an estimated final alcohol concentration. The method includes selecting, based on an output of the trained machine-learning model, a fermentation configuration that approximates the target organoleptic profile while maintaining the final alcohol concentration below the predefined threshold. The method also includes fermenting in accordance with the selected fermentation configuration to produce the fermented comestible having an alcohol content below the predefined threshold.

[0008] According to another aspect of the present disclosure, a fermented comestible produced according to the method described above is provided. The fermented comestible hasan alcohol content below the predefined threshold and exhibits an organoleptic profile approximating the target organoleptic profile.

[0009] According to a further aspect of the present disclosure, a computer-implemented method for generating a fermentation configuration for producing a fermented analog of an alcoholic beverage under an alcohol constraint is provided. The method includes receiving a target organoleptic profile derived from an alcoholic beverage. The method further includes applying a trained machine-learning model to evaluate candidate fermentation configurations by predicting, for each candidate fermentation configuration, an organoleptic profile relative to the target organoleptic profile and a final alcohol concentration. The method includes selecting, based on an output of the trained machine-learning model, a fermentation configuration predicted to approximate the target organoleptic profile while maintaining the predicted final alcohol concentration below a predefined threshold. The selected fermentation configuration is generated as a digital output for subsequent use, storage, or execution.

[0010] In yet another aspect of the present disclosure, a computer-implemented method for producing a fermented comestible analog of an alcoholic beverage having an alcohol content below a predefined threshold is provided. The method includes determining a target organoleptic profile of the alcoholic beverage. The method further includes applying a trained machine-learning model to evaluate candidate fermentation configurations by predicting, for each candidate fermentation configuration, an organoleptic profile relative to the target organoleptic profile and an estimated final alcohol concentration. The method includes selecting a fermentation configuration predicted to approximate the target organoleptic profile while maintaining the predicted final alcohol concentration below the predefined threshold. The method also includes fermenting in accordance with the selected fermentation configuration to produce the non-alcoholic comestible analog without post-fermentation alcohol removal.

[0011] According to still another aspect of the present disclosure, a system for producing a fermented comestible having an alcohol content below a predefined threshold is provided. The system includes one or more processors and a non -transitory computer-readable medium storing instructions that, when executed by the one or more processors, cause the system to receive a target organoleptic profile corresponding to desired flavor and aroma characteristics. The instructions further cause the system to apply a trained machine -learning model to evaluate candidate fermentation configurations by predicting, for each candidate fermentation configuration, an organoleptic profile and a final alcohol concentration. The instructions causethe system to select a fermentation configuration predicted to approximate the target organoleptic profile while maintaining the predicted final alcohol concentration below the predefined threshold. The instructions also cause the system to output fermentation execution data corresponding to the selected fermentation configuration for use in producing the fermented comestible without post-fermentation alcohol removal.

[0012] The foregoing general description of the illustrative embodiments and the following detailed description thereof are merely exemplary aspects of the teachings of this disclosure and are not restrictive.BRIEF DESCRIPTION OF FIGURES

[0013] Non-limiting and non-exhaustive examples are described with reference to the following figures.

[0014] FIG. 1 illustrates an example method for producing a non-alcoholic fermented beverage using a trained machine-learning model, according to aspects of the present disclosure.

[0015] FIG. 2 illustrates an example constrained multi -objective optimization performed using predictions generated by a trained machine-learning model for non-alcoholic fermentation, showing maximization of predicted organoleptic profile match while constraining predicted ethanol concentration, according to aspects of the present disclosure.

[0016] FIG. 3 illustrates examples of decision dimensions considered by the trained machine-learning model when generating and evaluating candidate fermentation configurations, according to aspects of the present disclosure.

[0017] FIG. 4 illustrates an example of a co-culture fermentation configuration that may be selected, sequenced, or evaluated by the trained machine -learning model.

[0018] FIG. 5 illustrates an example method for reverse engineering alcoholic beverage analogs using direct product fingerprinting and ingredient precursor deconstruction, according to aspects of the present disclosure.

[0019] FIG. 6 illustrates an example conceptual framework for recreating ethanol -associated sensory characteristics in a fermented analog, including parallel pathways for aroma replication, mouthfeel enhancement, and generation of warming sensory perception, according to aspects of the present disclosure.

[0020] FIG. 7 illustrates an example constrained multi-objective optimization process showing how a trained machine-learning model evaluates candidate fermentation configurations and generates predictions for selection, according to aspects of the present disclosure.

[0021] FIG. 8 illustrates example product characteristics of a non-alcoholic fermented comestible, including alcohol content, composition, shelf stability, and comestible types, according to aspects of the present disclosure.

[0022] FIG. 9 illustrates example fermentation process parameters 900 that may be included as variables within candidate fermentation configurations evaluated by the trained machine-learning model.DETAILED DESCRIPTION

[0023] The following description sets forth exemplary aspects of the present disclosure. It should be recognized, however, that such description is not intended as a limitation on the scope of the present disclosure. Rather, the description also encompasses combinations and modifications to those exemplary aspects described herein.

[0024] Overview: The present invention addresses challenges associated with producing non-alcoholic and zero-alcohol fermented comestibles that exhibit complex, adult-oriented flavor and aroma profiles comparable to traditional alcoholic comestible. As used herein, the term “fermented comestible” refers to a food or beverage product produced through fermentation and intended for human consumption. Fermented comestibles include liquid, semi-solid, and solid products, including but not limited to beverages, spoonable products, gels, pastes, or other ingestible matrices. References to fermented beverages herein are exemplary embodiments of fermented comestibles unless the context expressly requires otherwise.

[0025] In particular, the invention provides methods for producing a fermented comestible having a target organoleptic profile while maintaining final alcohol concentration below a defined threshold. In liquid embodiments, alcohol concentration may be expressed as alcohol by volume (ABV). In non-liquid embodiments, alcohol concentration may be expressed as an equivalent ethanol concentration by weight or other regulatory -recognized metric. As used herein, the term "non-alcoholic" refers to fermented comestibles having an alcohol content of less than or equal to 0.5% alcohol by volume (ABV), including trace amounts of ethanol permitted under applicable regulatory standards. The term "zero-alcohol" refers to fermentedcomestibles having an ethanol concentration below both the sensory detection threshold and the applicable regulatory threshold for labeling as containing alcohol in the relevant jurisdiction. In some embodiments, zero-alcohol corresponds to 0.0% ABV within analytical detection limits. Regulatory thresholds may differ between liquid and non-liquid comestibles and across jurisdictions. The disclosed methods avoid reliance on post-fermentation alcohol removal techniques and instead achieve alcohol control through fermentation design and execution. FIG. 1 illustrates an example method 100 for producing a non-alcoholic fermented comestible using a trained machine-learning model, according to aspects of the present disclosure.

[0026] Referring to FIG. 1, method 100 begins at step 102 with receiving a target organoleptic profile corresponding to desired flavor and aroma characteristics. At step 104, a trained machine-learning model is applied to solve a constrained optimization problem. The method proceeds in parallel to step 106, predicting an organoleptic profile, and step 108, predicting an alcohol concentration below a predefined threshold. At step 110, a substrate, strain or consortium, and fermentation parameters are selected based on the model output. At step 112, fermentation is performed in accordance with the selected configuration. At step 114, a non-alcoholic fermented comestible is produced.

[0027] The methods disclosed herein apply trained machine -learning models to guide decisions relating to fermentation inputs and conditions. The trained machine -learning models are used to evaluate candidate fermentation configurations 212 comprising candidate combinations of precursor substrates, microbial strains or microbial consortia, and fermentation process parameters under alcohol-constrained conditions. Fermentation is treated as the physical execution of decisions generated by the machine -learning model, rather than as an empirical or trial-and-error biological process.

[0028] This data-driven approach to fermentation design is referred to herein as “precision fermentation,” a term used to describe the precision of fermentation outcomes achieved through predictive modeling and constrained decision-making. As used in this disclosure, precision fermentation refers to the application of computational decision-making to fermentation design and does not require genetic engineering or modification of individual microbial strains. Instead, outcome precision is achieved through informed selection and coordination of naturally occurring, food-safe microorganisms, precursor substrates, and fermentation conditions.

[0029] As used herein, references to an organoleptic profile or alcohol concentration predicted by a trained machine-learning model refer to model-generated estimates corresponding to candidate fermentation configurations, and are distinct from measured properties of a fermented beverage produced by fermentation execution.

[0030] As used herein, a “fermentation configuration” refers to a defined set of fermentation inputs and conditions selected for execution, which may include, without limitation, a precursor substrate formulation, one or more food-safe microorganisms, and fermentation process parameters. The fermentation configuration represents the physical implementation of a candidate solution evaluated by the trained machine-learning model prior to fermentation execution

[0031] Constrained Multi-Objective Optimization: In certain embodiments, candidate fermentation configurations 212 predicted by the trained machine-learning model 702 to exceed a predefined alcohol threshold are excluded from further consideration prior to fermentation execution 706. In this manner, alcohol constraint is enforced at the design stage rather than corrected through post-fermentation processing. In certain embodiments, selection of microbial strains, precursor substrates, and fermentation process parameters is guided by application of a trained machine-learning model to a constrained, multi-objective optimization problem 200 specific to non-alcoholic fermentation. The trained machine-learning model is applied to fermentation outcome data describing relationships among substrate chemistry, microbial metabolic behavior, process parameters, and resulting fermentation products. Using this data, the model generates predictions relating to both organoleptic outcomes and ethanol production for candidate fermentation configurations.

[0032] A key aspect of the present disclosure is application of the trained machine-learning model 702 to address alcohol-constrained fermentation challenges directly, rather than optimizing fermentation outcomes without regard to alcohol formation. Candidate fermentation configurations 212 are evaluated against competing objectives, including predicted similarity to a target organoleptic profile 202 and predicted final ethanol concentration. As shown in FIG. 2, an optimization framework 200 applies organoleptic similarity objectives and alcohol constraints to guide evaluation of candidate configurations under non-alcoholic fermentation conditions. The predefined alcohol threshold may correspond to regulatory limits for non-alcoholic beverages, such as 0.5% ABV, or may correspond to stricter thresholds, including 0.0% ABV.

[0033] Candidate configurations predicted to exceed the threshold are excluded or penalized prior to fermentation execution 706. As a result, fermentation configurations are identified that are predicted to generate complex flavor and aroma profiles while limiting ethanol accumulation during fermentation. FIG. 7 illustrates an example implementation of this process 700, showing how the trained machine-learning model 702 evaluates candidate fermentation configurations 212 and produces outputs used for configuration selection 704 and subsequent fermentation execution 706.

[0034] Mechanisms for Alcohol Control: Mechanisms for alcohol control 300 are present. Specifically, ethanol production during fermentation is limited through several distinct but complementary control mechanisms that operate at different levels of the fermentation system. These control mechanisms address ethanol formation at the levels of microbial metabolism, substrate availability, and interspecies metabolic interaction. One or more of these mechanisms may be applied individually or in combination during fermentation design to achieve a target alcohol threshold.

[0035] FIG. 3 illustrates example mechanisms for alcohol control that may be evaluated and selected by a trained machine-learning model during generation of candidate fermentation configurations. As shown, these mechanisms include strain selection for non-ethanolic metabolic pathways 302, substrate design to limit fermentable sugars 304, and co-culture ethanol metabolism 306. Rather than biological inventions themselves, these mechanisms are controllable dimensions considered by the machine -learning model when predicting fermentation outcomes under alcohol-constrained conditions.

[0036] In a first control mechanism, ethanol production may be limited through microbial strain selection 302. Microbial strains may be selected based on metabolic characteristics associated with reduced ethanol formation relative to conventional ethanologenic yeasts. Such strains preferentially direct carbon flux toward metabolites other than ethanol, including glycerol, which contributes to mouthfeel, organic acids that contribute tartness and flavor complexity, and cellular biomass. Non-limiting examples include non-Saccharomyces yeasts, lactic acid bacteria, acetic acid bacteria, and other food-safe microorganisms exhibiting weakly ethanologenic or non-ethanologenic behavior. Selection of such strains reduces ethanol accumulation while preserving or enhancing flavor complexity.

[0037] In a second control mechanism, ethanol production is limited through precursor substrate formulation 304. The precursor substrate formulation is selected to restrict theavailability of fermentable sugars that would otherwise drive ethanologenic pathways. In some embodiments, fermentable monosaccharides are present at a concentration of less than about 10%. In other embodiments, the precursor substrate includes carbohydrates that are weakly fermentable or non-fermentable by ethanologenic microorganisms, including dextrins, oligosaccharides, pentoses, non-fermentable fibers, or combinations thereof. At the same time, the precursor substrate provides alternative nutrients and molecular precursors that support metabolic pathways associated with aroma and flavor development. By controlling substrate composition prior to fermentation, ethanol formation is constrained from the outset of the process.

[0038] In a third control mechanism, ethanol concentration is reduced through co -culture metabolism 306. In certain embodiments, a microbial consortium is employed that includes a first strain 404 that produces flavor-active metabolites and, in some cases, trace amounts of ethanol 408, and a second strain 410 that consumes ethanol produced by the first strain. The second strain thereby reduces ethanol concentration during fermentation while contributing secondary metabolites, such as organic acids, that enhance sensory complexity.

[0039] For example, FIG. 4 illustrates a co-culture fermentation configuration 400 in which a yeast strain such as Pichia kluyveri may generate flavor-active metabolites (here, fruity esters) 406 and trace ethanol 408, while a second microbial strain 410 such asAcetobacter aceti in the same consortium consumes the ethanol 408 produced by the first strain as a carbon source and converts it to acetic acid 412. FIG. 4 is provided for illustrative purposes to demonstrate an example fermentation configuration that may be evaluated and selected by the trained machine-learning model, and is not intended to limit the invention to any particular biological pathway, organism, or metabolic mechanism.

[0040] Precursor Substrate Design: Turning now to precursor substrate design 304, the precursor substrate 402 supports low-alcohol fermentation while supporting development of complex flavor and aroma profiles. In the present disclosure, precursor substrates are selected such that substrates commonly associated with high ethanol production are reduced or absent. In particular, the precursor substrate formulation may lack sugars that are readily fermented by high-ethanol-producing yeasts, thereby reducing the potential for significant alcohol formation at the outset of fermentation.

[0041] In some embodiments, the precursor substrate formulation limits fermentable monosaccharides. For example, fermentable monosaccharides may be present at aconcentration of less than about 10% of total fermentable carbohydrate content of the precursor substrate formulation. In other embodiments, the precursor substrate includes carbohydrates that are weakly fermentable or non-fermentable by ethanologenic microorganisms, including dextrins, oligosaccharides, pentoses, non-fermentable fibers, or combinations thereof. These components may contribute to mouthfeel or serve as substrates for non -ethanol metabolic pathways without substantially increasing ethanol production.

[0042] In certain embodiments, the precursor substrate formulation is substantially free of malted grain or other grain-derived substrates commonly associated with alcoholic fermentation. Instead, the precursor substrate may include one or more non -traditional feedstocks 514, such as fruit juices, teas, spices, herbs, hops, upcycled ingredients, alternative sugar sources, or combinations thereof. Upcycled ingredients may include spent grain, fruit pomace, vegetable byproducts, or imperfect produce. Alternative sugar sources may include monk fruit extract, stevia, allulose, or other rare sugars. These feedstocks provide chemical diversity and flavor-precursor components rather than serving primarily as fermentable sugar sources, and may be non-fermentable or weakly fermentable while remaining relevant to organoleptic outcomes evaluated by the trained machine -learning model.

[0043] In some embodiments, a trained machine-learning model 702 evaluates candidate precursor substrates to selectnon-traditional substrates 514 containing molecular precursors 512 corresponding to a target organoleptic profile 202. Such molecular precursors may include terpenes, phenolic compounds, thiol precursors, amino acids, or other compounds capable of being biotransformed into volatile aroma compounds during fermentation. For example, herbs such as coriander may provide precursors for linalool, while teas may provide polyphenolic or aromatic precursors. Selection of such precursor substrates enables generation of desirable aroma and flavor compounds during fermentation while limiting fermentable sugar availability that would otherwise promote ethanol production.

[0044] Microbial Strain Selection: Next, microbial strains used in the fermentation processes described herein are selected based on functional and role-based characteristics relevant to low-alcohol fermentation. In particular, selected strains are food-safe microorganisms suitable for use in fermented beverages, such as organisms recognized as generally regarded as safe (GRAS) or included on qualified presumption of safety (QPS) lists. Strain selection is based on metabolic behavior and functional contribution to fermentation outcomes rather than limitation to specific genera or species.

[0045] In certain embodiments, microbial strains are selected based on metabolic pathways associated with reduced ethanol formation. Such strains preferentially convert available sugars into metabolites other than ethanol, including glycerol, which contributes to mouthfeel, organic acids that contribute tartness and flavor complexity, and cellular biomass. Selection based on these functional characteristics enables suppression of ethanol accumulation while supporting development of complex flavor and aroma profiles.

[0046] In some embodiments, microbial strains are employed in the form of a microbial consortium comprising multiple strains with complementary metabolic roles. For example, a first strain may produce flavor-active metabolites, such as esters or other volatile compounds, and may generate trace amounts of ethanol during fermentation. A second strain included in the same consortium may consume ethanol produced by the first strain as a carbon source, thereby reducing ethanol concentration during fermentation. In a non-limiting example, a yeast strain such as Pichia kluyveri may generate fruity esters and trace ethanol, while an acetic acid bacterium such as Acetobacter aceti consumes ethanol and converts it to acetic acid. This cooperative metabolic interaction supports concurrent flavor development and ethanol suppression.

[0047] Reverse Engineering of Alcoholic Beverage Analogs: A key application of the methods described herein is the reverse engineering of existing products, such as commercial alcoholic beverages or other fermented products, to define a target organoleptic profile 202 for a non-alcoholic analog. FIG. 5 illustrates an example reverse-engineering method 500 in which a target alcoholic beverage 502 is analyzed using one or both of two complementary pathways. In a first pathway, direct product fingerprinting 504 is performed using analytical chemistry techniques 506 to generate a chemical fingerprint 508. In a second pathway, ingredient precursor deconstruction 510 is used to identify molecular precursors 512 associated with flavor, aroma, or mouthfeel attributes. Based on the identified molecular precursors, one or more non-traditional substrates 514 may be selected. Outputs of the reverse-engineering process are used to generate a target organoleptic profile 202, which may be evaluated or refined using a trained machine-learning model 702 and provided as an input to subsequent fermentation design under alcohol-constrained conditions.

[0048] In the first pathway 504, a target alcoholic beverage 502 is directly analyzed using analytical chemistry techniques 506. Non-limiting examples of such techniques include gas chromatography-mass spectrometry (GC-MS), high-performance liquid chromatography (HPLC), or related analytical methods. These analyses generate a quantitative chemicalfingerprint 508 of the target product, identifying volatile organic compounds, esters, phenolic compounds, acids, and other molecules that contribute to aroma, flavor, and mouthfeel. The resulting fingerprint may be used to define a target organoleptic profile 202 in terms of measurable chemical features.

[0049] In the complementary pathway 510, the traditional ingredients of the target product are deconstructed into their constituent molecular precursors 512 via an ingredient precursor deconstruction pathway 510. Ingredients such as malt, hops, fruits, or botanicals may be analyzed to identify compounds or precursor molecules responsible for characteristic sensory attributes. A trained machine-learning model 702 may then be applied to evaluate candidatenon-traditional substrates 514 that contain corresponding molecular precursors. In this manner, alternative substrates may be selected that provide similar flavor -relevant chemistry while avoiding substrates commonly associated with high ethanol production.

[0050] The reverse-engineered target organoleptic profile 202 obtained using one or both of these methods may then be used as an input to the fermentation design process. Using this target profile, candidate fermentation configurations 212 comprising combinations of precursor substrates, microbial strains or consortia, and fermentation parameters are evaluated to identify configurations predicted to approximate the target organoleptic profile while maintaining final alcohol concentration below a defined threshold.

[0051] Following selection of precursor substrates and microbial strains or consortia, fermentation is carried out under controlled process parameters selected to support development of the target organoleptic profile while limiting ethanol formation. These parameters govern the physical execution of fermentation and operate in concert with substrate composition and microbial metabolism to achieve low-alcohol outcomes. FIG. 9 illustrates example fermentation process parameters 900 that may be included as variables within candidate fermentation configurations evaluated by the trained machine-learning model, including oxygen availability 902, temperature control 904, pH control 906, and staged nutrient availability 908.

[0052] In some embodiments, fermentation temperature 904 is selected to favor production of flavor-active metabolites over ethanol. Temperature ranges may be chosen outside ranges typically associated with peak ethanologenic activity of conventional alcoholic fermentation yeasts. Adjusting fermentation temperature 904 may influence enzyme activity,metabolic flux, and growth kinetics of the selected microorganisms, thereby contributing to reduced ethanol accumulation.

[0053] In certain embodiments, fermentation pH control 906 is controlled within a range that supports microbial activity and flavor development while limiting ethanol formation. For example, pH may be maintained within a mildly acidic range conducive to organic acid production and microbial stability. pH control 906 may also contribute to shelf stability of the resulting fermented beverage.

[0054] In some embodiments, oxygen availability 902 during fermentation is controlled to influence metabolic pathways of the selected microorganisms. Fermentation may be conducted under aerobic, micro-aerobic, or oxygen-limited conditions depending on the metabolic characteristics of the microbial strains employed. Modulating oxygen availability 902 may reduce ethanol formation while promoting production of desirable aroma and flavor compounds.

[0055] In further embodiments, staged nutrient availability 908 is staged or limited during fermentation. For example, nitrogen sources, vitamins, or trace nutrients may be supplied in a controlled manner to influence microbial growth and metabolite production. Staged nutrient availability 908 may reduce excessive sugar consumption and redirect metabolic activity toward non-ethanol pathways.

[0056] Fermentation duration may also be selected based on predicted organoleptic outcomes and ethanol accumulation. In some embodiments, fermentation is terminated upon achieving a target flavor and aroma profile and prior to accumulation of ethanol above a defined threshold. Termination may include cooling, removal of microorganisms, or other suitable process steps known in the art. The combination of these fermentation process parameters enables controlled execution of fermentation designs selected using machine-learning-guided evaluation, resulting in fermented beverages that approximate target organoleptic profiles while maintaining alcohol content below regulatory or desired thresholds.

[0057] Recreation of Ethanol Sensory Experience: A key application of the present invention is the production of fermented non-alcoholic comestibles, including beverages, that exhibit sensory characteristics commonly associated with distilled spirits. In addition to replicating flavor and aroma, the methods described herein enable evaluation and selection of fermentation designs that reproduce physical sensory attributes associated with ethanol -containing beverages, including mouthfeel and warming sensations, while maintaining alcoholcontent below a defined threshold. As conceptually illustrated in FIG. 6, the recreation of ethanol sensory experience 600 may involve evaluation of three parallel sensory pathways, including aroma replication 602 through volatile organic compounds 604, mouthfeel enhancement 606 through viscosity-increasing metabolites such as glycerol and polysaccharides 608, and warming sensation 610 through trigeminal -stimulating compounds 612.

[0058] In some embodiments, aroma characteristics associated with distilled spirits are approximated through fermentation-derived volatile compounds 604. Fermentation configurations may be selected or evaluated under conditions predicted to result in formation of volatile organic compounds characteristic of aged or barrel-stored products. Such compounds may include esters, phenolic compounds, aldehydes, and other aroma -active molecules. For example, when producing a whiskey analog, fermentation may be predicted to result in formation of compounds such as vanillin, selected esters, and phenolic compounds that contribute to aromas commonly associated with oak aging. A trained machine -learning model 702 may be applied to evaluate and rank candidate fermentation configurations for their predicted ability to produce target aroma profiles.

[0059] In further embodiments, mouthfeel and body characteristics associated with ethanol-containing beverages are reproduced through fermentation-derived metabolites via the mouthfeel pathway 606. Ethanol contributes to viscosity and perceived body in distilled spirits; in the present invention, similar sensory attributes are predicted and achieved through production of texture-enhancing metabolites during fermentation. In some embodiments, fermentation configurations are selected that are predicted to result in increased levels of glycerol and selected polysaccharides 608, including exopolysaccharides, which increase viscosity and contribute to a smoother mouthfeel. Selection of microbial strains, precursor substrates, and fermentation parameters may be guided by predictions generated by the trained machine-learning model 702 based on metabolite profiles associated with enhanced body and texture.

[0060] In additional embodiments, a warming or spicy sensory perception commonly associated with ethanol consumption is produced without ethanol through the warming sensation pathway 610 and formation of alternative sensory -active compounds. In certain embodiments, fermentation configurations are evaluated for their predicted production or retention of trigeminal-stimulating compounds 612 that contribute to a warming or spicy perception on the palate. Such compounds may include phenolic compounds, spice -derivedmolecules, or other fermentation-derived constituents associated with trigeminal sensory perception. For example, fermentation may be designed using precursor materials associated with compounds such as eugenol or ginger-derived components, resulting in comestibles that exhibit a warming sensory character without reliance on ethanol. The resulting spirit analog 614 may combine one or more of these sensory pathways to approximate the sensory experience of an ethanol-containing beverage.

[0061] Product Characteristics and Shelf Stability: In some embodiments, the fermented comestible produced according to the methods described herein is a beverage or food product characterized by a complex organoleptic profile and a reduced residual sugar content. The fermented comestible has an alcohol content of less than 0.5% alcohol by volume (ABV), and in certain embodiments, 0.0% ABV. FIG. 8 illustrates example product characteristics 800 of a non-alcoholic fermented comestible, including alcohol content, composition, shelf stability, and comestible types.

[0062] In certain embodiments, the fermented comestible exhibits shelf stability at room temperature for an extended period, such as at least six months, without requiring pasteurization or the addition of synthetic preservatives. Shelf stability may arise from a combination of factors including controlled fermentation conditions, reduced residual fermentable sugars, production of organic acids, and establishment of an unfavorable environment for spoilage organisms. In still other embodiments, the fermented comestible is suitable for distribution and storage under ambient conditions without refrigeration. The fermented comestible may include beverages such as beer, wine, cider, and spirit analogs, as well as food products such as yogurt-style products, cheese-style products, and other fermented food items.

[0063] For example, in a first mechanism, ethanol production is limited through microbial strain selection. The machine-learning model selects microbial strains or microbial consortia that preferentially convert available sugars into metabolites other than ethanol. Such metabolites may include glycerol, which enhances mouthfeel, organic acids that contribute tartness and flavor complexity, and cellular biomass. In some embodiments, the selected microbial strains exhibit metabolic behavior in which carbon flux is directed away from ethanologenic pathways. Non-limiting examples include non-Saccharomyces yeasts whose fermentation metabolism produces minimal ethanol relative to conventional alcoholic fermentation yeasts. Selection of such strains reduces ethanol formation while supporting production of flavor-active compounds.

[0064] In a second mechanism, ethanol production is limited through precursor substrate formulation. The precursor substrate formulation is selected to restrict the availability of fermentable sugars that would otherwise drive ethanologenic pathways. In some embodiments, the precursor substrate formulation comprises fermentable monosaccharides at a concentration of less than about 10%. In other embodiments, the precursor substrate formulation comprises carbohydrates that are weakly fermentable or non -fermentable by ethanologenic microorganisms, including dextrins, oligosaccharides, pentoses, non -fermentable fibers, or combinations thereof. By limiting fermentable sugar availability while providing alternative nutrients and flavor precursors, ethanol formation is constrained from the outset of fermentation.

[0065] In a third mechanism, ethanol concentration is reduced through co -culture metabolism. In certain embodiments, a microbial consortium is employed that includes a first strain that produces flavor-active metabolites and, in some cases, trace amounts of ethanol, and a second strain that consumes ethanol produced by the first strain. The second strain thereby reduces ethanol concentration during fermentation while contributing additional metabolites that enhance sensory complexity. For example, a yeast strain such as Pichia kluyveri may generate fruity esters and trace ethanol, while an acetic acid bacterium such as Acetobacter aceti, included in the same consortium, consumes the ethanol as a carbon source and converts it to acetic acid. This cooperative metabolic interaction enables concurrent flavor development and ethanol suppression.

[0066] Unlike conventional empirical fermentation approaches, the methods described herein rely on a trained machine-learning model 702 that encodes fermentation-specific relationships between precursor substrate chemistry, microbial strain behavior, fermentation process parameters, and resulting organoleptic and alcohol outcomes. The trained model enables prediction of fermentation results prior to fermentation execution, thereby allowing fermentation designs to be evaluated and selected under alcohol-constrained conditions. In the methods described herein, the trained machine-learning model is not a generic predictive tool, but a fermentation-specific model trained on data describing relationships among precursor substrate composition, microbial strain behavior, fermentation process parameters, and resulting fermentation outcomes. Such outcomes include measurable organoleptic attributes and ethanol concentration. By capturing these relationships, the trained model enables prediction of fermentation results prior to fermentation execution 706 and allows candidate fermentation configurations 212 to be evaluated and selected via a selection process 704 underalcohol-constrained conditions. This predictive capability permits fermentation to be conducted as a planned, model-guided process rather than as an empirical or trial -and-error activity.

[0067] A number of implementations have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the disclosure. Accordingly, other implementations are within the scope of the following claims.

Claims

CLAIMS1. A computer-implemented method for producing a fermented comestible having an alcohol content below a predefined threshold, the method comprising:(a) receiving a target organoleptic profile corresponding to desired flavor and aroma characteristics;(b) applying a trained machine-learning model to evaluate candidate fermentation configurations, the trained machine-learning model generating for each candidate fermentation configuration:(i) an organoleptic profile, and (ii) an estimated final alcohol concentration;(c) selecting, based on an output of the trained machine-learning model, a fermentation configuration that approximates the target organoleptic profile while maintaining the final alcohol concentration below the predefined threshold; and(d) fermenting in accordance with the selected fermentation configuration to produce the fermented comestible,wherein the fermented comestible has an alcohol content below the predefined threshold.

2. The method of claim 1, wherein the fermented comestible is produced without postfermentation alcohol removal.

3. The method of claim 1, wherein each candidate fermentation configuration comprises a precursor substrate formulation, one or more food-safe microorganisms, and fermentation process parameters.

4. The method of claim 1, wherein the one or more food-safe microorganisms comprise a microbial consortium.

5. The method of claim 1, wherein the trained machine -learning model is trained on fermentation outcome data capturing relationships between fermentation inputs and resulting metabolite concentrations.

6. The method of claim 1, wherein the trained machine -learning model is trained on fermentation outcome data generated from fermentations conducted under alcohol-constrained conditions.

7. The method of claim 1, wherein candidate fermentation configurations predicted to result in a final alcohol concentration exceeding the predefined threshold are excluded prior to selection.

8. The method of claim 1, wherein applying the trained machine-learning model comprises evaluating candidate fermentation configurations under competing objectives including predicted organoleptic similarity and predicted alcohol concentration.

9. The method of claim 1, wherein the trained machine -learning model ranks candidate fermentation configurations based on predicted proximity to the target organoleptic profile subject to the predefined threshold.

10. The method of claim 1, wherein the trained machine-learning model generates the organoleptic profile and the final alcohol concentration for each candidate fermentation configuration prior to fermentation execution.

11. The method of claim 1, wherein the trained machine-learning model predicts metabolite production arising from interactions among members of a microbial consortium that differ from metabolite production by individual members in isolation.

12. The method of claim 1, wherein the one or more food-safe microorganisms comprise a plurality of microorganisms introduced sequentially or in staged inoculation steps, and wherein the trained machine-learning model predicts metabolite production resulting from metabolism of fermentation byproducts produced by an earlier-inoculated microorganism.

13. The method of claim 12, wherein the fermentation byproducts comprise ethanol produced during an earlier fermentation stage.

14. The method of claim 1, wherein the trained machine-learning model evaluates candidate fermentation configurations based on predicted net fermentation outcomes that cannot be determined from individual inputs in isolation.

15. The method of claim 1, wherein the predefined threshold corresponds to 0.5% alcohol by volume for beverages or a corresponding ethanol content by weight for solids.

16. The method of claim 1, wherein the predefined threshold corresponds to 0.0% alcohol by volume for beverages or a corresponding ethanol content by weight for solids.

17. A fermented comestible produced according to the method of claim 1, wherein the fermented comestible has an alcohol content below the predefined threshold and exhibits an organoleptic profile approximating the target organoleptic profile.

18. A computer-implemented method for generating a fermentation configuration for producing a fermented analog of an alcoholic beverage under an alcohol constraint, the method comprising:(a) receiving a target organoleptic profile derived from an alcoholic beverage;(b) applying a trained machine-learning model to evaluate candidate fermentation configurations by predicting, for each candidate fermentation configuration: (i) an organoleptic profile relative to the target organoleptic profile, and (ii) a final alcohol concentration; and(c) selecting, based on an output of the trained machine-learning model, a fermentation configuration predicted to approximate the target organoleptic profile while maintaining the predicted final alcohol concentration below a predefined threshold,wherein the selected fermentation configuration is generated as a digital output for subsequent use, storage, or execution.

19. A computer-implemented method for producing a fermented comestible analog of an alcoholic beverage having an alcohol content below a predefined threshold, comprising: (a) determining a target organoleptic profile of the alcoholic beverage;(b) applying a trained machine-learning model to evaluate candidate fermentation configurations by predicting, for each candidate fermentation configuration: (i) an organoleptic profile relative to the target organoleptic profile, and (ii) an estimated final alcohol concentration;(c) selecting a fermentation configuration predicted to approximate the target organoleptic profile while maintaining the predicted final alcohol concentration below the predefined threshold; and(d) fermenting in accordance with the selected fermentation configuration to produce the nonalcoholic comestible analog,wherein the non-alcoholic comestible analog is produced without post-fermentation alcohol removal.

20. The method of claim 19, wherein determining the target organoleptic profile comprises analyzing an alcoholic beverage using one or more analytical chemistry techniques to generatea chemical fingerprint representing volatile aroma compounds, flavor -active metabolites, mouthfeel-contributing compounds, or combinations thereof.

21. The method of claim 20, wherein the analytical chemistry techniques comprise gas chromatography-mass spectrometry (GC-MS), high-performance liquid chromatography (HPLC), or combinations thereof.

22. The method of claim 19, wherein determining the target organoleptic profile comprises deconstructing traditional ingredients of the alcoholic beverage into constituent molecular precursors associated with aroma, flavor, or mouthfeel attributes.

23. The method of claim 19, wherein applying the trained machine-learning model comprises evaluating candidate precursor substrate formulations based on predicted biotransformation of molecular precursors into fermentation-derived aroma compounds.

24. The method of claim 19, wherein applying the trained machine-learning model comprises predicting interaction effects between candidate precursor substrate formulations and one or more food-safe microorganisms that influence metabolite production under alcohol-constrained conditions.

25. The method of claim 19, wherein candidate fermentation configurations predicted to result in an alcohol concentration exceeding the predefined threshold are excluded from selection prior to fermentation execution.

26. The method of claim 19, wherein the predefined threshold corresponds to a non-alcoholic limit of 0.5% alcohol by volume.

27. The method of claim 19, wherein the predefined threshold corresponds to 0.0% alcohol by volume.

28. The method of claim 19, wherein the trained machine -learning model evaluates candidate fermentation configurations to approximate sensory attributes selected from aroma, flavor complexity, mouthfeel, or combinations thereof, without reliance on ethanol.

29. The method of claim 19, wherein the non-alcoholic comestible analog is an analog of beer, wine, cider, a distilled spirit, or a fermented food product.

30. The method of claim 19, wherein fermentation is terminated based on a prediction generated by the trained machine-learning model indicating achievement of the target organoleptic profile prior to predicted accumulation of ethanol above the predefined threshold.

31. The method of claim 19, wherein the trained machine -learning model is applied using fermentation outcome data generated from alcohol-constrained fermentation runs.

32. The method of claim 19, wherein the trained machine -learning model predicts shelf stability of the non-alcoholic analog based on predicted organic acid production, residual sugar content, and pH.

33. The method of claim 29, wherein the non-alcoholic analog of a distilled spirit comprises fermentation-derived compounds predicted by the trained machine -learning model to approximate:(i) aroma characteristics associated with barrel aging;(ii) mouthfeel characteristics associated with ethanol viscosity; and(iii) warming sensory perception associated with ethanol consumption.

34. A system for producing a fermented comestible having an alcohol content below a predefined threshold, comprising:(a) one or more processors; and(b) a non-transitory computer-readable medium storing instructions that, when executed by the one or more processors, cause the system to:(i) receive a target organoleptic profile corresponding to desired flavor and aroma characteristics;(ii) apply a trained machine-learning model to evaluate candidate fermentation configurations by predicting, for each candidate fermentation configuration, an organoleptic profile and a final alcohol concentration;(iii) select a fermentation configuration predicted to approximate the target organoleptic profile while maintaining the predicted final alcohol concentration below the predefined threshold; and(iv) output fermentation execution data corresponding to the selected fermentation configuration for use in producing the fermented comestible without post-fermentation alcohol removal.

Images