Method and system for producing fermented comestibles with targeted metabolite profiles

Abstract

A computer-implemented method for producing a fermented comestible having a target physiological or functional effect is provided. The method involves receiving a target input corresponding to a desired physiological or functional effect, including mood modulation, cognitive focus, or sleep support. A trained machine-learning model identifies a candidate combination including at least one precursor substrate, at least one food-safe microbial strain or microbial consortium, and associated fermentation process parameters. The trained machine-learning model predicts production of two or more metabolites associated with the desired physiological or functional effect at concentrations within a target metabolite profile. The precursor substrate, microbial strain or consortium, and fermentation process parameters are selected based on the model output. The selected precursor substrate is fermented with the selected microbial strain or consortium under the selected fermentation process parameters to produce the fermented comestible having metabolites at concentrations corresponding to the target physiological or functional effect.

Description

METHOD AND SYSTEM FOR PRODUCING FERMENTED COMESTIBLES WITH TARGETED METABOLITE PROFILESCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Patent Application No.63 / 740,952, titled SYSTEM AND METHOD FOR OPTIMIZING CLEAN ALCOHOLIC BEVERAGE FERMENTATION PROCESSES BASED ON ARTIFICIAL INTELLIGENCE MACHINE LEARNING MODELS, filed December 31, 2024, and to U.S. Provisional Patent Application No. 63 / 740,956, titled SYSTEM AND METHOD FOR OPTIMIZING MOOD BEVERAGE FERMENTATION PROCESSES BASED ON ARTIFICIAL INTELLIGENCE MACHINE LEARNING MODELS, which are hereby incorporated by reference in their 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 machine learning systems for fermentation, and more particularly, to a system and method for designing microbial consortia with targeted metabolite profiles for physiological effects.BACKGROUND

[0003] Humans have long sought comestibles that provide sustained energy, acute cognitive focus, stress resilience, relaxation, social enjoyment, or other mood -associated effects. Traditional options have included various supplements, such as orally consumed nootropic and adaptogenic supplements, alcoholic beverages or the like. While alcoholic beverages may provide transient relaxation or social facilitation, they are also associated with well-documented drawbacks, including impairment, undesirable byproducts, and variability in composition. In addition, many commercially available alcoholic products may conflict with modern wellness preferences, often containing high residual sugar, allergenic substrates, or industrial additives. Nootropic and adaptogenic supplements or other supplements, by contrast, are typically consumed in isolated form. However, these may also be limited by delayed onset,inconsistent effects, or poor consumer acceptance. Additionally, many such ingredients possess strong bitter or earthy flavor profiles which can limit palatability.

[0004] Fermentation has long been used to produce beverages and food products with complex flavor profiles. However, traditional fermentation processes, particularly those used for beverages such as kombucha or foods such as yogurt, often yield variable outcomes. Therefore, flavor, alcohol content, residual sugar, and metabolite composition may also fluctuate. Attempts to address these limitations have included post-fermentation processing techniques, such as alcohol removal or blending. However, these can negatively affect sensory characteristics and increase manufacturing complexity. Functional ingredients such as adaptogens and nootropics present additional formulation challenges. Active compounds are often embedded within complex plant matrices, and their incorporation into consumable products frequently results in undesirable sensory characteristics or inconsistent functional profiles.

[0005] Accordingly, a need exists to addresses some of these and 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] In one aspect of the present disclosure, a computer-implemented method for producing a fermented comestible having a targeted physiological or functional effect is provided. The method includes receiving a target input corresponding to a desired physiological or functional effect. The method involves using a trained machine-learning model to identify a candidate combination including at least one precursor substrate, at least one food-safe microbial strain or microbial consortium, and associated fermentation process parameters. The trained machine-learning model is configured to predict production of two or more metabolites associated with the desired physiological or functional effect and to predict concentrations of the two or more metabolites within a predefined target range. The method includes selecting the precursor substrate, the microbial strain or microbial consortium, and the fermentation process parameters based on an output of the trained machine -learning model. The method further includes fermenting the selected precursor substrate with the selectedmicrobial strain or microbial consortium in accordance with the selected fermentation process parameters to produce the fermented comestible.

[0008] In an aspect of the present disclosure, the computer-implemented method for producing a fermented comestible is configured to modulate mood is provided. The method includes receiving a target input indicative of a desired mood-related physiological effect. The method involves applying a trained machine -learning model to identify one or more precursor substrates including molecular precursors associated with mood-related metabolites. The method further involves applying the trained machine -learning model to select at least one microbial strain or microbial consortium predicted to biotransform the molecular precursors into the mood-related metabolites under one or more fermentation process parameters. The method includes fermenting the precursor substrates with the selected microbial strain or microbial consortium under the one or more fermentation process parameters to produce the fermented comestible. The fermented comestible includes two or more mood -related metabolites at concentrations corresponding to the desired mood-related physiological effect.

[0009] In yet another aspect of the present disclosure, a fermented comestible produced according to the disclosed methods is provided. The fermented comestible includes two or more fermentation-derived metabolites present at relative concentrations corresponding to a target physiological or functional effect selected during fermentation design.

[0010] In still another aspect of the present disclosure, a system for producing a fermented comestible having a targeted physiological or functional effect is provided. The system includes one or more processors and non -transitory computer-readable memory storing instructions that, when executed by the one or more processors, cause the system to receive a target input corresponding to a desired physiological or functional effect. The instructions cause the system to apply a trained machine-learning model configured to predict production of two or more metabolites associated with the desired physiological or functional effect and to predict concentrations of the two or more metabolites within a predefined target range. The trained machine-learning model evaluates candidate combinations including at least one precursor substrate, at least one food-safe microbial strain or microbial consortium, and associated fermentation process parameters. The instructions cause the system to select at least one precursor substrate, at least one microbial strain or microbial consortium, and associated fermentation process parameters based on an output of the trained machine -learning model. The instructions cause the system to output fermentation instructions for fermenting theselected precursor substrate with the selected microbial strain or microbial consortium under the selected fermentation process parameters to produce the fermented comestible.

[0011] In a further aspect of the present disclosure, a computer-implemented method for producing a fermented comestible having a target physiological or functional effect is provided. The method includes receiving a target input corresponding to a desired physiological or functional effect. The method involves using a trained machine-learning model to identify a candidate combination including at least one precursor substrate, at least one food-safe microbial strain or microbial consortium, and associated fermentation process parameters that produces two or more metabolites at concentrations within a predetermined tolerance of a target metabolite profile associated with the desired physiological or functional effect. The method includes selecting the candidate combination based on an output of the trained machine-learning model. The method further includes fermenting the selected precursor substrate with the selected microbial strain or microbial consortium in accordance with the selected fermentation process parameters to produce the fermented comestible.

[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 fermented comestible having a targeted physiological or functional effect using a trained machine -learning model, according to aspects of the present disclosure.

[0015] FIG. 2 illustrates an example method for producing a fermented comestible configured to modulate mood, showing biotransformation of molecular precursors into mood-related metabolites, according to aspects of the present disclosure.

[0016] FIG. 3 illustrates an example mapping of target inputs corresponding to desired physiological or functional effects to associated metabolite targets, according to aspects of the present disclosure.

[0017] FIG. 4 illustrates an example system architecture for producing a fermented comestible, including processors, memory, and instructions for applying a trained machine -learning model and outputting fermentation instructions, according to aspects of the present disclosure.

[0018] FIG. 5 illustrates an example biotransformation pathway showing transformation of adaptogen-derived compounds during fermentation, including reduction of bitter compounds, according to aspects of the present disclosure.

[0019] FIG. 6 illustrates an example co-production of ethanol and functional metabolites in a single fermentation process, according to aspects of the present disclosure.

[0020] FIG. 7 illustrates example product characteristics of a fermented comestible, including metabolite profile, functional components, composition, and shelf stability, according to aspects of the present disclosure.

[0021] FIG. 8 illustrates an example method for matching a target metabolite profile, showing evaluation of candidate combinations to produce metabolites within a predetermined tolerance of the target profile, according to aspects of the present disclosure.

[0022] FIG. 9 illustrates an example training data structure for a machine-learning model, showing microbial strain genomic features, substrate molecular precursor features, and measured post-fermentation metabolite concentrations, according to aspects of the present disclosure.

[0023] FIG. 10 illustrates an example comparison of consortium metabolite production versus individual strain characteristics, showing strain-strain interactions and complementary metabolic activities that influence metabolite production, according to aspects of the present disclosure.

[0024] FIG. 11 illustrates an example iterative model refinement loop, showing measurement of metabolite concentrations, computation of residuals, and updating of machine -learning model parameters, according to aspects of the present disclosure.

[0025] FIG 12A illustrates an example parallel fermentation process for concentration amplification, showing a primary fermentor and a parallel fermentor operating concurrently with controlled blending, according to aspects of the present disclosure.

[0026] FIG 12B illustrates an example sequential fermentation process for functional -to-sensory layering, showing a primary fermentor producing a functional -rich intermediate and a secondary fermentor for aroma and texture refinement, according to aspects of the present disclosure.

[0027] FIG 12C illustrates an example single-fermentor process with temporal substrate addition, showing multiple phases including base fermentation, functional push, and sensory finish within a single vessel, according to aspects of the present disclosure.DETAILED DESCRIPTION

[0028] 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, but rather encompasses combinations, substitutions, and modifications of the exemplary aspects described herein, as will be appreciated by persons of ordinary skill in the art.

[0029] In one aspect of the present disclosure, a computer-implemented method 100 for producing a fermented comestible having a targeted physiological or functional effect 104 is provided. The targeted physiological effect 104 may be, for example, mood modulation, sustained energy, acute cognitive focus, stress resilience, or another physiological effect. The method 100 includes receiving a target input 102 corresponding to a desired physiological or functional effect 104. The method involves using a trained machine-learning model 107 to identify a candidate combination 108 including at least one precursor substrate 122, at least one food-safe microbial strain or microbial consortium 124, and associated fermentation process parameters 126. The trained machine -learning model 107 is configured to predict production of two or more metabolites 132 associated with the desired physiological or functional effect and to predict concentrations of the two or more metabolites within a predefined target range 134. The method includes selecting the precursor substrate, the microbial strain or microbial consortium, and the fermentation process parameters 110 based on an output of the trained machine-learning model. The method further includes fermenting 112 the selected precursor substrate with the selected microbial strain or microbial consortium in accordance with the selected fermentation process parameters to produce the fermented comestible 116.

[0030] In various embodiments, the target input 102 may be received through different input mechanisms. For example, the target input 102 may be received through a graphical user interface 220 presenting selectable physiological or functional effects. Alternatively, the target input 102 may be received through a natural-language interface 222 accepting text-based queries describing desired effects. In yet other implementations, the target input 102 is receivedthrough an application programming interface (API) 224 receiving programmatic requests from external systems. The target input 102 may also be selected from a predefined library of effects or generated based on user preferences, historical usage data, or combinations thereof, as will be apparent to a person skilled in the art.

[0031] Upon receiving a target input 102, a trained machine-learning model 107 is applied to evaluate candidate fermentation inputs. The trained model 107 is trained on fermentation outcome data 140 comprising historical fermentation runs. Such fermentation outcome data 140 includes, by way of example and without limitation, precursor substrate compositions, microbial strain identities and inoculation rates, fermentation process parameters, and measured metabolite concentrations in resulting fermented products. By capturing relationships between fermentation inputs and metabolite outputs, the training data enables the model to predict metabolite production for candidate combinations that were not directly observed during training. The internal architecture and training methodology of the machinelearning model are described in the incorporated PCT application.

[0032] According to certain implementations, microbial strain features used to train the machine-learning model 107 comprise genomic and functional descriptors. Such features may include gene presence or absence indicators, metabolic pathway annotations, predicted enzyme activities, or strain-specific metabolic capabilities. Substrate molecular precursor features (144) may include amino acid content, phenolic compound profiles, fermentable sugar composition, adaptogen-derived precursor concentrations, or other chemical or molecular descriptors of precursor substrates. Together, these features enable the trained machinelearning model (107) to generate predictions for previously unobserved combinations of precursor substrates (122), microbial strains or microbial consortia (124), and fermentation process parameters (126).

[0033] In further embodiments, candidate fermentation inputs comprise microbial consortia (124) including two or more microbial strains. When microbial consortia are employed, the consortium may exhibit strain-strain interactions (174), cross-feeding effects (176), or complementary metabolic activities (178) that influence metabolite production in ways not predictable from individual strain characteristics evaluated in isolation. By way of example, a first strain (170) may produce metabolic intermediates that serve as substrates for a second strain (172), resulting in metabolite profiles that differ from those achievable by either strain alone.

[0034] To address this technical challenge, the trained machine -learning model (107) is configured to learn interaction-dependent metabolite production from fermentation outcome data (140) generated using multi-strain consortium fermentations. In some embodiments, the fermentation outcome data (140) comprises paired pre -fermentation substrate profiles and post-fermentation metabolite profiles obtained from consortium fermentations. This approach enables the model to capture emergent metabolic behaviors arising from complementary or interacting metabolic pathways operating in concert, thereby improving prediction accuracy for multi-strain fermentation designs relative to approaches based solely on individual strain characteristics.

[0035] In some embodiments, a precursor substrate composition is selected by the trained machine-learning model (107) to function as a metabolic modulator that induces pathway activation, intermediate production, or metabolite transformation by a single microbial strain, such that the resulting metabolite profile approximates effects otherwise achievable through microbial consortia. By way of example, substrate precursors may activate dormant biosynthetic pathways, induce production of metabolic intermediates, or trigger enzymatic transformations that would otherwise require complementary metabolic activities from multiple strains.

[0036] Turning now to FIG. 4, an example system architecture for producing a fermented comestible is illustrated. The system (400) includes one or more processors (402) and non-transitory computer-readable memory (404) storing instructions (406). The present disclosure relates to systems and methods for producing fermented comestibles (116) using a computer-implemented design framework. The disclosure leverages a machine -learning platform for designing fermentation processes, as described in PCT Application No. PCT / IB2025 / 059740, titled SYSTEM AND METHOD FOR PRESCRIBING FERMENTATION BASED ON MACHINE LEARNING MODELS (filed September 27, 2025), the contents of which are incorporated herein by reference in their entirety. The instructions (406), when executed by the one or more processors (402), cause the system (400) to receive a target input (102), apply the trained machine-learning model (107), evaluate candidate combinations (108), select (110) a combination based on the model output, and output (114) fermentation instructions (115).

[0037] With reference to FIG. 8, an example method for matching a target metabolite profile (130) is illustrated. In the present disclosure, the platform is applied at an application and execution level to design fermentation processes that produce fermented comestibles (116) having targeted metabolite profiles (130) corresponding to desired physiological or functionaleffects. Such effects include, by way of example and without limitation, mood modulation, cognitive focus, sleep support, gut health, metabolic health, immune support, and antiinflammatory response.

[0038] FIG. 1 illustrates an example method (100) for producing a fermented comestible (116) having a targeted physiological or functional effect. According to certain embodiments, the method begins by receiving a target input (102) corresponding to a desired physiological or functional effect (104). The target input (104) may be provided by a user through a graphical or natural-language interface, selected from a predefined library of effects, or generated programmatically based on other system inputs. Target inputs include, without limitation: mood modulation (e.g., relaxation, alertness, stress reduction, emotional balance), cognitive focus, sleep support, gut health, metabolic health, immune support, and anti-inflammatory response.

[0039] Each target input (104) is associated with a target metabolite profile (130) representing target concentrations, concentration ranges, or relative levels of two or more metabolites (132). The mapping between target inputs and metabolite profiles may be implemented using lookup tables, rulesets, trained classifiers, ontologies, or other techniques known to a skilled person, without requiring explicit modeling of physiological signaling mechanisms. The trained model (107) predicts (109) production of two or more metabolites (132) at concentrations within the predefined target range (134). Based on the prediction, a combination is selected (110) that most closely aligns with the target metabolite profile (130). The following examples help illustrate these mappings.

[0040] Example 1: "Relaxation" Target Input Target metabolite profile: GABA: 50-150 mg / L Tryptophan: 20-80 mg / L Acetate: 300-1000 mg / L Ethanol: <0.5% v / v

[0041] Example 2: "Alertness" Target Input Target metabolite profile: Tyrosine: 40-120 mg / L Propionate: 200-800 mg / L Glutamate: 75-250 mg / L These examples are illustrative and non-limiting; alternative metabolites, ranges, or combinations may be used depending on the desired physiological or functional effect.

[0042] FIG. 2 illustrates an example method for producing a fermented comestible (116) configured to modulate mood. In some embodiments, the method includes receiving a target input (102) indicative of a desired mood-related physiological effect. The method then involves applying (106) a trained machine-learning model (107) to identify one or more precursor substrates (122) including molecular precursors associated with mood-related metabolites(138). The method further involves applying (108) the trained model (107) to select at least one microbial strain or microbial consortium (124) predicted to biotransform the molecular precursors into the mood-related metabolites (138) under one or more fermentation process parameters (126). Subsequently, the method includes fermenting (112) the precursor substrates (122) with the selected microbial strain or microbial consortium (124) under the one or more fermentation process parameters (126) to produce the fermented comestible (116). The resulting fermented comestible (116) includes two or more mood-related metabolites (138) at concentrations corresponding to the desired mood-related physiological effect.

[0043] The following example illustrates application of the disclosed method (100) to produce a fermented comestible (116) targeting relaxation. This example is illustrative and not limiting.

[0044] A target input (104) corresponding to relaxation is received (102). The target input (104) is associated with a target metabolite profile (130) including GABA at approximately 50-150 mg / L, tryptophan at approximately 20-80 mg / L, and acetate at approximately 300-1000 mg / L.

[0045] The trained model (107) evaluates (108) candidate combinations (120) and identifies a candidate combination (120) predicted to produce metabolites (132) within the target ranges. In this example, the selected candidate combination (120) includes: a precursor substrate (122) comprising ashwagandha extract at approximately 2% w / v in an apple juice base; a microbial consortium (124) comprising Lactobacillus plantarum at approximately 107CFU / mL and Hanseniaspora uvarum at approximately 106CFU / mL; and fermentation process parameters (126) including temperature of approximately 22°C, pH of approximately 4.2, and fermentation duration of approximately 72 hours.

[0046] The precursor substrate (122) is then fermented (112) with the selected microbial consortium (124) under the selected fermentation process parameters (126). The resulting fermented comestible (116) includes GABA at approximately 92 mg / L, tryptophan at approximately 45 mg / L, and acetate at approximately 620 mg / L, corresponding to the target metabolite profile (130) for relaxation.

[0047] Turning to FIG. 3, an example mapping of target inputs (104) to associated metabolite targets is illustrated. Having defined a target metabolite profile (130), a trained machine-learning model (107) is used to evaluate candidate fermentation combinations (120). The trained model (107) has been trained on fermentation outcome data (140) capturingrelationships between precursor substrate composition, microbial strain selection, fermentation process parameters, and resulting metabolite concentrations, as detailed in the incorporated PCT application.

[0048] The trained model (107) evaluates (108) candidate combinations (120) comprising: at least one precursor substrate (122), at least one food-safe microbial strain or microbial consortium (124), and associated fermentation process parameters (126). The model determines whether such combinations will produce two or more metabolites (132) within predefined target ranges (134) or within a predetermined tolerance (136) of a target metabolite profile (130). In certain implementations, the model iteratively evaluates multiple candidate combinations (120) and identifies one or more combinations predicted to most closely align with the target metabolite profile (130) according to similarity or agreement criteria, such as distance metrics or percentage deviation. Such evaluation may be implemented using any suitable technique described in the incorporated reference or otherwise known to skilled persons.

[0049] Example 3: Model-Assisted Selection Workflow Input: "Relaxation" target Topranked output: Substrate: Ashwagandha (2% w / v) Microbes: Lactobacillus plantarum (107CFU / mL) + Hanseniaspora uvarum (106CFU / mL) Parameters: 22 °C, pH 4.2, 72 hours Predicted outcome: GABA: 92 mg / L Tryptophan: 45 mg / L Acetate: 620 mg / L

[0050] With reference to FIG. 9, an example training data structure for a machine-learning model (107) is illustrated. In some embodiments, the trained model (107) is trained on fermentation outcome data (140) comprising at least three categories of features. The first category includes microbial strain genomic features (142), including, by way of example and without limitation, metabolic pathway annotations, strain -specific metabolite production profiles, predicted enzyme activities, or other genomic or functional features known to a skilled person. The second category involves substrate molecular precursor features (144), including amino acid content, phenolic compound profiles, fermentable sugar composition, adaptogen-derived precursor concentrations, or related chemical descriptors. The third category involves measured post-fermentation metabolite concentrations (146) from prior fermentation processes. Collectively, the training data captures relationships between these fermentation inputs and resulting metabolite outputs, enabling the model to predict multi-dimensional metabolite profiles for previously unobserved combinations of precursor substrates (122), microbial strains or microbial consortia (124), and fermentation process parameters (126).

[0051] Turning to FIG. 11, an example iterative model refinement loop is illustrated. In some embodiments, after fermentation execution, the method further comprises measuring concentrations (190) of the two or more metabolites (132) in the produced fermented comestible (116). A residual (193) is then computed (192) between the measured concentrations (191) and concentrations predicted by the trained model (107). Based on this residual (193), the parameters of the trained model (107) are updated (194) using the measured concentrations (191) and computed residual (193) to improve prediction accuracy for subsequent fermentation designs. In certain implementations, updating (194) comprises retraining or adjusting parameters of the trained model (107) using the computed residual (193) to reduce prediction error for subsequent fermentation designs involving similar precursor substrate (122) and microbial strain or consortium (124) combinations. Through this iterative refinement, continuous improvement of prediction accuracy is achieved as additional fermentation outcome data (140) is collected.

[0052] In some embodiments, fermentation outcome data used for iterative refinement of the trained model further comprises measured outputs beyond metabolite concentration data. Such measured outputs may include, by way of example and without limitation, validated physiological metrics, biomarker measurements, functional response data, or structured sensory evaluation data associated with consumption or assessment of the fermented comestible.

[0053] In such embodiments, the trained model may incorporate these measured outputs as additional training signals or validation inputs when updating model parameters, thereby improving prediction of fermentation designs associated with desired physiological or functional effects. Use of such measured outputs does not require clinical diagnosis or treatment and may rely on non-invasive, population-level, or aggregate data, as will be understood by a person skilled in the art. Such empirical efficacy data may be incorporated into the feedback loop illustrated in FIG. 11.

[0054] In some embodiments, the trained model (107) identifies candidate combinations (120) that produce two or more metabolites (132) at concentrations within a predetermined tolerance (136) of a target metabolite profile (130) associated with the desired physiological or functional effect. The predetermined tolerance (136) may be expressed as a percentage deviation, an absolute concentration range, or a distance metric in a multi-dimensional metabolite space. The model iteratively evaluates multiple candidate combinations (120) andidentifies one or more combinations predicted to most closely align with the target metabolite profile (130) according to similarity or agreement criteria.

[0055] With reference to FIG. 5, an example biotransformation pathway is illustrated showing transformation of adaptogen-derived compounds during fermentation. In some embodiments, the precursor substrate (122) includes adaptogenic plant materials (150) containing bitter compounds (152) such as withanolides, steroidal saponins, or phenolic glycosides. During fermentation, the microbial strain or consortium (124) may perform deglycosylation (154) of withanolides, hydrolysis (156) of saponins, and decarboxylation (158) of phenolic compounds, resulting in transformed compounds (160) with reduced bitterness, reduced astringency, increased bioaccessibility, and improved palatability.

[0056] In some embodiments, fermentation of botanical, adaptogenic, or nootropic precursor substrates results in chemical or structural modification of one or more phytochemicals present in the precursor substrate. Such modification may alter bioavailability, solubility, stability, or functional performance of the phytochemicals relative to their form in the unfermented substrate.

[0057] By way of example and without limitation, such modification may include changes in molecular conjugation state, polarity, molecular association with fermentation -derived metabolites, or matrix-level interactions that affect absorption or functional activity. These modifications may arise from microbial metabolism, fermentation conditions, or co-produced metabolites, without requiring any particular enzymatic mechanism or specific chemical transformation.

[0058] Accordingly, references herein to biotransformation of phytochemicals are intended to encompass, without limitation, any fermentation-induced chemical, structural, or matrix-level modification that alters functional characteristics of the phytochemicals, as will be appreciated by a person skilled in the art.

[0059] As used herein, food-safe microbial strains include strains having Generally Recognized as Safe (GRAS) status under U.S. Food and Drug Administration regulations, Qualified Presumption of Safety (QPS) status under European Food Safety Authority guidelines, or equivalent regulatory approval for use in food or beverage production in other jurisdictions. Food-safe microbial strains may also include strains with a history of safe use in traditional fermented foods and beverages. When identifying candidate combinations (120), the trained model (107) evaluates candidate strains from a library of food-safe strains.

[0060] FIG. 6 illustrates an example co-production of ethanol (162) and functional metabolites (132) in a single fermentation process. Selected combinations (120) are fermented (112) without post-fermentation blending of separately manufactured metabolites. The single fermentation process may co-produce ethanol (162) within a predefined range, postbiotic metabolites (164) including SCFAs, organic acids, and vitamins, and maintain probiotic organisms (166) at viable counts of at least 106CFU / mL. Fermentation may result in modification of molecular or matrix environments of functional compounds, thereby increasing stability or bioaccessibility, without requiring complete degradation of precursor substrates (122).

[0061] In some embodiments, ethanol is produced during fermentation as one component of a multi-metabolite profile selected by the trained model (107). The concentration of ethanol may be selected as a design variable based on regulatory, sensory, or functional objectives and is not a defining limitation of the disclosed methods. Ethanol concentrations may therefore fall within non-intoxicating ranges, intoxicating ranges, or intermediate ranges, depending on the target metabolite profile (130) and intended use of the fermented comestible. As used herein, "non-intoxicating threshold" refers to ethanol concentrations at or below approximately 0.5% v / v, or such other threshold as defined by applicable food and beverage regulatory standards in a given jurisdiction.

[0062] In some embodiments, the fermentation process produces a plurality of metabolites that are co-present within a shared fermented matrix. Rather than acting as isolated components, such co-produced metabolites may collectively influence stability, solubility, bioaccessibility, sensory perception, or physiological performance of one or more functional compounds present in the fermented comestible. For example, organic acids, short-chain fatty acids, amino acids, or fermentation-derived cofactors may modify the physicochemical environment of other metabolites, including neurotransmitter precursors or adaptogen -derived compounds, thereby influencing their dispersion, persistence, or uptake. These effects arise from co-production during fermentation without post-fermentation blending or addition of separately manufactured components.

[0063] In some embodiments, the fermented comestible comprises two or more fermentation-derived metabolites that exhibit synergistic physiological or functional effects. As used herein, "synergistic" refers to a combined effect arising from the co -presence of two or more metabolites that is different from, enhanced relative to, or not directly predictable from the effects of the metabolites individually when evaluated in isolation. Such synergistic effectsmay include, without limitation, increased intensity, duration, or qualitative differences in a physiological or functional response relative to individual metabolites administered separately.

[0064] By way of example and without limitation, a fermented comestible comprising gamma-aminobutyric acid (GABA), one or more short-chain fatty acids, and trace ethanol produced during a single fermentation process may elicit a relaxation or stress -reduction response that differs from, or is enhanced relative to, the responses associated with administration of equivalent quantities of GABA, short-chain fatty acids, or ethanol individually. In some embodiments, such effects may be observed in sensory, functional, or physiological assessments of the fermented comestible.

[0065] In some embodiments, the trained model (107) is configured to evaluate multimetabolite profiles and predict fermentation outcomes associated with combined physiological or functional effects. The model may be trained on fermentation outcome data (140) reflecting observed responses to metabolite combinations, without requiring explicit modeling of biochemical mechanisms underlying such combined effects. Candidate combinations (120) may be selected based on predicted alignment with target profiles (130) associated with such combined or interaction-dependent effects, including those arising from co-production of metabolites within a fermented matrix.

[0066] As used herein, "co-fermentation" or "co-produced" refers to production of two or more metabolites as part of a coordinated fermentation process designed according to a common fermentation design output, regardless of whether such fermentation occurs within a single fermentation vessel, across multiple fermentation vessels operating in parallel, or across multiple fermentation stages executed sequentially. Co-fermentation may therefore include simultaneous fermentation in separate vessels, staged fermentation across time, or combinations thereof, provided that the resulting metabolites are produced according to a unified fermentation design and combined to form the fermented comestible.

[0067] With reference to FIG. 12C, in some embodiments, the fermentation process is executed within a single fermentation vessel (602), wherein one or more microbial strains or consortia act concurrently on one or more precursor substrates to produce the fermented comestible. The single-vessel fermentation may include multiple temporal phases, including a base fermentation phase (604), a functional push phase (606) in which target precursors are added to activate specific metabolic pathways, and a sensory finish phase (608) in which aroma substrates are added to modulate ester, thiol, or mouthfeel characteristics.

[0068] With reference to FIG. 12 A, in other embodiments, fermentation is executed using parallel fermentation (610), wherein two or more fermentation vessels operate concurrently under coordinated fermentation parameters. A primary fermentor (612) may produce a base matrix including postbiotic metabolites, SCFAs, organic acids, and baseline aroma compounds. A parallel fermentor (614) may be optimized for production of a concentrated functional metabolite stream (616). Metabolites produced in parallel vessels may be combined via controlled blending (618) during or after fermentation to form the fermented comestible having a target functional load and balanced sensory profile.

[0069] With reference to FIG. 12B, in yet other embodiments, fermentation is executed using sequential fermentation (620), wherein fermentation occurs in two or more stages. In a first stage, a primary fermentor (622) produces a functional -rich intermediate (624) with target bioactive metabolites and controlled ethanol suppression. In a subsequent stage, a secondary fermentor (626) performs aroma, texture, and biotransformation processes including ester production, thiol release, and polysaccharide or glycerol modulation. The resulting fermented comestible comprises an integrated functional and sensory matrix.

[0070] In each of the foregoing embodiments, the trained model (107) is used to select fermentation inputs and process parameters such that the resulting fermented comestible (116) exhibits a target metabolite profile (130) or functional effect, irrespective of the specific execution mode.

[0071] Resulting fermented comestibles (116) may exhibit several characteristics, including: two or more metabolites (132) at concentrations corresponding to target physiological or functional effects; compositional stability at room temperature; viable probiotics (166) (e.g., >106CFU / mL) or postbiotic metabolites (164) such as SCFAs, organic acids, or vitamins; and unpasteurized, naturally derived flavor with reduced residual sugar.

[0072] Turning to FIG. 7, example product characteristics of a fermented comestible (116) are illustrated. The fermented comestible (116) may be characterized by its metabolite profile (702), functional components (704), composition (706), and shelf stability (708). Fermentation (112) may be conducted using fermentation vessels with controlled temperature, pH monitoring, and oxygen management capabilities. In some embodiments, the fermentation vessels include sensors for monitoring fermentation progress and controllers for adjusting process parameters. The fermentation instructions (115) output by the system (400) may be executed by automated fermentation equipment configured to receive and implement thespecified parameters. Alternatively, the fermentation instructions (115) may be provided to operators who configure fermentation equipment and monitor fermentation progress according to the specified parameters. Fermentation duration may range from approximately 24 hours to approximately 120 hours depending on the selected microbial strains and target metabolite profile (130).

[0073] Shelf stability of the fermented comestible (116) may result from one or more fermentation-derived characteristics such as reduced residual sugars, acidic pH, antimicrobial metabolites or others. In some embodiments, fermentation produces organic acids that reduce pH to levels inhibitory to spoilage microorganisms, such as pH below 4.0. In other embodiments, fermentation results in production of antimicrobial metabolites, including bacteriocins, organic acids, or other compounds that inhibit growth of undesirable microorganisms. In yet other embodiments, substantially complete fermentation of available sugars reduces residual sugar content to levels that do not support growth of spoilage organisms. Additionally, competitive exclusion by beneficial microorganisms present in the fermented comestible (116) may further contribute to shelf stability. These mechanisms may operate individually or in combination to achieve shelf stability without heat pasteurization or synthetic preservatives.

[0074] Metabolite concentrations in the predefined target range (134) are selected such that a single serving of the fermented comestible (116) delivers a quantity of functional metabolites (132) corresponding to the desired physiological or functional effect. In some embodiments, a single serving is approximately 100 to 355 mL. The target concentrations account for expected bioavailability of the metabolites and typical consumption patterns. For example, a relaxation-targeted beverage may be formulated such that a single 250 mL serving provides GABA, tryptophan, and short-chain fatty acids at concentrations associated with relaxation effects based on available scientific literature or empirical data.

Claims

CLAIMS1. A computer-implemented method for producing a fermented comestible having a targeted physiological or functional effect, the method comprising:(a) receiving a target input corresponding to a desired physiological or functional effect; (b) using a trained machine-learning model to identify a candidate combination comprising:(i) at least one precursor substrate,(ii) at least one food-safe microbial strain or microbial consortium, and(iii) associated fermentation process parameters,wherein the trained machine-learning model is configured to predict production of two or more metabolites associated with the desired physiological or functional effect and to predict concentrations of the two or more metabolites within a predefined target range;(c) selecting the precursor substrate, the microbial strain or microbial consortium, and the fermentation process parameters based on an output of the trained machine-learning model; and(d) fermenting the selected precursor substrate with the selected microbial strain or microbial consortium in accordance with the selected fermentation process parameters to produce the fermented comestible.

2. The method of claim 1 , wherein the desired physiological or functional effect is selected from mood modulation, cognitive focus, sleep support, gut health, metabolic health, immune support, or anti-inflammatory response.

3. A computer-implemented method for producing a fermented comestible configured to modulate mood, the method comprising:(a) receiving a target input indicative of a desired mood-related physiological effect; (b) applying a trained machine-learning model to identify one or more precursor substrates comprising molecular precursors associated with mood-related metabolites; (c) applying the trained machine-learning model to select at least one microbial strain or microbial consortium predicted to biotransform the molecular precursors into the mood-related metabolites under one or more fermentation process parameters; and (d) fermenting the precursor substrates with the selected microbial strain or microbial consortium under the one or more fermentation process parameters to produce thefermented comestible,wherein the fermented comestible comprises two or more mood-related metabolites at concentrations corresponding to the desired mood-related physiological effect.

4. The method of claim 3, wherein the mood-related physiological effect is selected from relaxation, alertness, stress reduction, or emotional balance.

5. The method of claim 1, wherein the two or more metabolites comprise at least one neurotransmitter precursor.

6. The method of claim 5, wherein the neurotransmitter precursor is selected from GABA or GABA precursors, serotonin precursors, dopamine precursors, tryptophan, tyrosine, or combinations thereof.

7. The method of claim 1, wherein the two or more metabolites further comprise one or more short-chain fatty acids selected from acetate, propionate, or butyrate.

8. The method of claim 1, wherein the predefined target range specifies approximate concentration ranges including:GABA at approximately 10-200 mg / L,glutamate at approximately 50-500 mg / L,tryptophan at approximately 10-100 mg / L,tyrosine at approximately 10-150 mg / L, andtotal short-chain fatty acids at approximately 200-2000 mg / L.

9. The method of claim 1, wherein the precursor substrate comprises one or more botanical, adaptogenic, or nootropic materials.

10. The method of claim 9, wherein the precursor substrate comprises materials derived from Lion’s mane (Hericium erinaceus), Ashwagandha (Withania somnifera), Rhodiola rosea, Bacopa monnieri, Ginkgo biloba, or combinations thereof.

11. The method of claim 1, wherein ethanol is one of the metabolites predicted by the trained machine-learning model and is maintained within a predefined non-intoxicating concentration range.

12. The method of claim 11, wherein the fermented comestible has an ethanol concentration below 0.5% by volume.

13. The method of claim 1, wherein the fermented comestible is produced without postfermentation blending of separately manufactured metabolites.

14. The method of claim 1 , wherein the fermentation results in modification of a molecular or matrix environment of at least one functional compound such that the functional compound exhibits increased chemical stability or bioaccessibility relative to an unfermented precursor substrate.

15. The method of claim 1, wherein the fermented comestible has a residual sugar content below a predetermined threshold selected to support shelf stability without heat pasteurization.

16. The method of claim 1 , wherein a single serving of the fermented comestible comprises approximately 100 to 355 mL and includes a fermented functional matrix containing at least one neurotransmitter precursor and at least one postbiotic metabolite.

17. The method of claim 1, wherein the trained machine -learning model is trained on fermentation outcome data comprising relationships between precursor substrate composition, microbial strain selection, fermentation process parameters, and final metabolite concentrations.

18. A fermented comestible produced according to the method of claim 1,wherein the fermented comestible comprises two or more fermentation -derived metabolites present at relative concentrations corresponding to a target physiological or functional effect selected during fermentation design.

19. The fermented comestible of claim 18, wherein the target physiological or functional effect comprises a mood-related physiological effect and the two or more fermentation- derived metabolites comprise at least one neurotransmitter precursor and at least one short-chain fatty acid.

20. The fermented comestible of claim 18, wherein the fermented comestible maintains extended compositional stability at room temperature comparable to pasteurized products.

21. The fermented comestible of claim 18, wherein the fermented comestible comprises:(a) a viable probiotic microorganism present at or above a defined minimum viable count, or(b) a postbiotic metabolite selected from organic acids, short-chain fatty acids, or vitamins,or combinations thereof.

22. The fermented comestible of claim 18, wherein the fermented comestible is unpasteurized, and flavor and aroma characteristics are derived from fermentation of natural precursor substrates without post-fermentation addition of synthetic flavors or additives, and the fermented comestible has a residual sugar content below a predetermined threshold.

23. A system for producing a fermented comestible having a targeted physiological or functional effect, the system comprising:one or more processors; andnon-transitory computer-readable memory storing instructions that, when executed by the one or more processors, cause the system to:receive a target input corresponding to a desired physiological or functional effect; apply a trained machine-learning model configured to predict production of two or more metabolites associated with the desired physiological or functional effect and to predict concentrations of the two or more metabolites within a predefined target range, to evaluate candidate combinations comprising at least one precursor substrate, at least one food-safe microbial strain or microbial consortium, and associated fermentation process parameters;select at least one precursor substrate, at least one microbial strain or microbial consortium, and associated fermentation process parameters based on an output of the trained machine-learning model; andoutput fermentation instructions for fermenting the selected precursor substrate with the selected microbial strain or microbial consortium under the selected fermentation process parameters to produce the fermented comestible.

24. The system of claim 23, wherein the instructions, when executed, further cause the system to receive fermentation outcome data, determine an observed metabolite profile of a fermented comestible, and update parameters of the trained machine-learning model based on a difference between the observed metabolite profile and a target metabolite profile associated with the target input.

25. The system of claim 23, wherein the target input corresponds to a mood -related physiological effect, and the predefined target range specifies target concentrations or relative levels of two or more mood-associated metabolites including at least one neurotransmitter precursor and at least one short-chain fatty acid.

26. A computer-implemented method for producing a fermented comestible having a target physiological or functional effect, the method comprising:(a) receiving a target input corresponding to a desired physiological or functional effect; (b) using a trained machine-learning model to identify a candidate combination comprising at least one precursor substrate, at least one food -safe microbial strain or microbial consortium, and associated fermentation process parameters that produces two or more metabolites at concentrations within a predetermined tolerance of a target metabolite profile associated with the desired physiological or functional effect;(c) selecting the candidate combination based on an output of the trained machine - learning model; and(d) fermenting the selected precursor substrate with the selected microbial strain or microbial consortium in accordance with the selected fermentation process parameters to produce the fermented comestible.

27. The fermented comestible of claim 18, wherein ethanol is present at a concentration above a non-intoxicating threshold and is co-produced with one or more postbiotic metabolites during a single fermentation process without post-fermentation blending.

28. The method of claim 1, wherein:(a) the trained machine-learning model is trained on fermentation outcome data comprising:(i) microbial strain genomic features,(ii) substrate molecular precursor features, and(iii) measured post-fermentation metabolite concentrations from prior fermentation processes; and(b) the trained machine-learning model is configured to predict multi-dimensional metabolite profiles for previously unobserved combinations of precursor substrates, microbial strains or microbial consortia, and fermentation process parameters.

29. The method of claim 1 , wherein:(a) the microbial consortium exhibits strain-strain interactions, cross-feeding effects, or complementary metabolic activities that influence metabolite production in a manner not predictable from individual strain characteristics in isolation; and(b) the trained machine-learning model is configured to predict such interactiondependent metabolite production by learning from fermentation outcome datacomprising paired pre-fermentation substrate profiles and post-fermentation metabolite profiles generated from microbial consortium fermentations.

30. The method of claim 1, further comprising:(a) after step (d), measuring concentrations of the two or more metabolites in the produced fermented comestible;(b) computing a residual between the measured concentrations and concentrations predicted by the trained machine-learning model; and(c) updating parameters of the trained machine -learning model using the measured concentrations and computed residual to improve prediction accuracy for subsequent fermentation designs.

31. The method of claim 28, wherein the microbial strain genomic features comprise gene presence / absence matrices and metabolic pathway annotations derived from whole genome sequences.

32. The method of claim 29, wherein the trained machine-learning model learns strainstrain interactions from experimental fermentation outcome data comprising multi - strain consortium fermentations in which metabolite production differs from that of corresponding single-strain fermentations due to cross-feeding or complementary metabolic activity.

33. The method of claim 30, wherein updating comprises retraining or adjusting parameters of the trained machine-learning model using the computed residual to reduce prediction error for subsequent fermentation designs involving similar precursor substrate and microbial strain or consortium combinations.

34. The method of claim 30, wherein updating parameters of the trained machine-learning model is further based on measured outputs associated with the fermented comestible, the measured outputs comprising measured metabolite concentrations, analytical measurements, physiological response data, biomarker data, biometric data, sensory assessment data, or combinations thereof.

35. The method of claim 1, wherein the two or more metabolites are co-produced during fermentation and exhibit interaction-dependent effects arising from their co-presence within the fermented comestible, such that the resulting physiological or functional effect is not predictable from individual metabolite concentrations evaluated in isolation.

36. The method of claim 35, wherein the co-produced metabolites exhibit functional effects such that the combined effect of the two or more metabolites differs from an effect associated with administration of the metabolites individually.

37. The method of claim 1, wherein fermenting comprises executing fermentation in parallel, sequential, or coordinated multi-vessel processes under coordinated fermentation parameters.

Images