The human gut contains over 1,000 bacterial species, each responding differently to dozens of prebiotic compounds, each interacting with every other species in the community, each requiring a specific ratio to achieve a specific physiological outcome. The math exceeds what any expert, working alone, can reliably solve. This is exactly the kind of problem artificial intelligence was built for.
The formulation problem that conventional supplement science cannot solve
Here is the challenge facing anyone who wants to design a prebiotic that actually works with precision.
The human microbiome contains an estimated 38 trillion microorganisms across more than 1,000 species. Each species has distinct substrate preferences — the specific compounds it can ferment, and the metabolic outputs it produces when it does. Each species also exists in ecological relationship with every other species in the community: competing for resources, producing metabolites that affect neighboring populations, contributing to a pH environment that either supports or suppresses other organisms.
Now add this: different physiological goals require different microbial targets. A person whose primary concern is supporting healthy sleep quality has a different microbial profile pattern than someone focused on metabolic wellness or cardiovascular support. The bacterial communities that need nourishment, and the specific ratios in which they need to be nourished, differ meaningfully between these goals.
And then add the final layer of complexity: the prebiotic compounds themselves interact with each other. The ratio between inulin and arabinoxylan oligosaccharides affects which species gain the competitive advantage. The presence of quercetin modulates how Akkermansia responds to other substrates in the formula. These interaction effects multiply across every additional compound in the blend.
The honest assessment: A skilled human formulator working with five or ten prebiotic compounds can make educated, experience-informed decisions. But optimizing across dozens of compounds, hundreds of bacterial targets, thousands of interaction effects, and multiple physiological goal profiles simultaneously — and arriving at precise milligram-level ratios for each — is beyond the reliable capacity of human intuition. It is a computational problem.
What AI was actually built for
There is a great deal of noise around artificial intelligence in the wellness and supplement space. Most of it amounts to a chatbot answering questions or a quiz generating a generic recommendation. That is not what we are describing here.
The AI system underlying MAIOME's formulation is a genuine machine learning pipeline — three interconnected neural network modules trained on the largest microbiome database in the world, cross-referenced with the world's only dedicated microorganism-nutrient interaction database. It was developed by a research team that built the architecture from the ground up specifically for microbiome analysis and modulation.
This is the kind of problem AI is legitimately suited to solve: a high-dimensional optimization challenge with real biological data, validated against measurable outcomes, operating at a resolution no human analysis can match.
We want to be direct about why this matters — not just for MAIOME's formulas, but for the broader question of what AI should be used for. AI is a powerful tool. Its value is proportional to the integrity of the problem it is applied to and the quality of the data it is trained on. Applying a purpose-built, scientifically validated AI system to one of the most complex optimization problems in nutritional science is, in our view, exactly the right use of the technology. Not as a shortcut. As a capability amplifier — doing what human expertise alone cannot.
The three-module pipeline: how the system actually works
The AI operates as a sequential three-module pipeline. Each module handles a distinct analytical task, and the output of each feeds into the next. Understanding what each module does makes clear why no single algorithm — and no human formulator — could replicate what the full system produces.
Module 1: The taxonomic network — reading the microbiome
The first module processes raw DNA data from gut microbiome sequencing. Using a 1D convolutional neural network architecture — the same class of deep learning used in genomic analysis — it reads the DNA sequence data as a text corpus and constructs a taxonomic profile: a map of which bacterial species are present, in what abundances, and what the ecological relationships between them are.
This is not a simple species count. The taxonomic network identifies the structure of the microbial community — the hierarchical relationships between phyla, families, genera, and species — and uses those relationships to understand the ecosystem as a whole rather than as a list of individual organisms.
The network was trained using the Greengenes database, one of the most comprehensive 16S rRNA gene databases in existence, enabling accurate taxonomic classification across the full breadth of known gut bacterial diversity.
Module 2: The super resolution network — understanding what the microbiome is doing
The second module takes the taxonomic profile produced by module one and asks a deeper question: given this community of organisms, what is the microbiome actually doing? What metabolic pathways are active? What compounds is it producing? What health-relevant functions is it performing — or failing to perform?
This module uses a Generative Adversarial Network architecture with a proprietary PhyloNet structure — a biologically informed neural network that mirrors the phylogenetic relationships between bacterial species, allowing the model to make predictions that respect the actual evolutionary and ecological structure of the microbiome rather than treating species as independent variables.
The super resolution network was trained on a database of approximately 5,000 microbiome samples compiled from three of the world's largest open-source microbiome research projects — the American Gut Project, the Human Microbiome Project, and the Flemish Gut Flora Project — combined with proprietary in-house data. In total, the training database encompasses microbiome data and lifestyle and health information from over 33,000 individuals.
From this training, the network learned to predict 12 microbiome-associated health and wellness parameters, including those related to bowel health, metabolic markers, sleep quality, body composition, vitamin synthesis capacity, and the presence of beneficial probiotic organisms. All predictions were validated with cross-validation testing, achieving AUCROC scores above 0.8 across all parameters — a standard benchmark for clinically meaningful predictive accuracy.
The output of this module is a functional assessment of the microbiome: not just what organisms are present, but what the ecosystem is producing, what it is missing, and what physiological parameters it is affecting.
Why this matters: Most microbiome tests tell you what bacteria you have. This module tells you what your microbiome is doing with them — and what it would need to do differently to better support a specific health goal. That distinction is the difference between a species list and a functional map.
Module 3: The recommendation engine — computing the optimal formula
The third module is where the formulation happens. Given the functional assessment from module two — the specific microbial communities that need support, the metabolic pathways that need nourishment, the dysbiotic populations that need to be outcompeted — the recommendation engine computes the precise prebiotic intervention required to achieve the target modulation.
This module uses a self-attention neural network architecture, operating against a proprietary nutrient database that conventional nutrition databases cannot replicate. Standard nutrition databases track macronutrients, vitamins, and minerals. They do not contain the micronutrient and phytochemical data relevant to microbiome modulation at species-level resolution.
To build this database, the Enbiosis research team conducted a meta-analysis of over 3,000 peer-reviewed scientific publications and datasets — specifically mapping the relationships between individual food components and the specific bacterial species they modulate. The resulting database covers carbohydrates, lipids, proteins, vitamins and minerals, phytochemicals, specific fermented foods, and preservatives and additives — all cross-referenced to the microorganisms they influence and the direction of that influence.
The recommendation engine uses this database to compute the optimal composition of prebiotic compounds needed to achieve the specific modulation target — at milligram-level precision, for each formula goal.
The formulas this pipeline produced are available now. Each MAIOME formula is the direct computational output of this three-module system — precision-formulated for a specific physiological goal. [Explore the MAIOME Collection →]
From algorithm to formula: what precision actually means
The three-module pipeline produces something that has not previously existed in supplement formulation: a formula derived not from a formulator's best judgment about which ingredients are beneficial, but from a computational analysis of exactly which bacterial communities need specific substrates, in specific ratios, to achieve a specific and measurable modulation outcome.
The practical result is that each MAIOME formula corresponds to a distinct microbiome profile pattern associated with a specific physiological goal.
Someone whose primary focus is supporting healthy sleep quality presents a characteristic microbiome pattern — specific organisms underpresented, specific metabolic pathways underactive, specific bacterial communities that, when nourished, support the tryptophan routing toward serotonin and melatonin precursors that the gut-brain axis depends on. The formula for that goal contains the precise prebiotic compounds, at the precise milligram ratios, that the algorithm calculated would achieve that modulation.
Someone focused on metabolic wellness presents a different pattern — different species profiles, different functional deficits, different modulation targets. The formula is different. Not because a formulator made a different judgment call, but because the algorithm computed a different answer from the biological data.
The same logic applies across cardiovascular support, immune wellness, skin health, weight management, healthy aging, and the other goal categories in the MAIOME catalog. Each formula is the computational output of the same pipeline applied to a different input profile. The AI does not guess. It calculates.
The precision claim: When MAIOME uses the word precision, it has a specific meaning. It means milligram-level prebiotic ratios derived from a three-module AI pipeline trained on 33,000+ individual microbiome profiles, cross-referenced against a 3,000+ publication nutrient-microorganism interaction database, and validated against measurable health parameter predictions. That is what the word earns.
The integrity question: when should AI be used?
We think it is worth being direct about something that gets lost in most AI-driven product conversations: using AI responsibly means using it for problems where its capabilities genuinely exceed what human expertise can provide, and where the data it operates on is rigorous enough to trust the output.
The microbiome formulation problem meets both criteria. The complexity of the optimization — dozens of prebiotic compounds, hundreds of bacterial targets, thousands of interaction effects, multiple physiological goal profiles — genuinely exceeds what human expertise can reliably navigate at milligram-level precision. And the training data — 33,000+ real human microbiome profiles, validated against measurable health outcomes, drawn from the world's largest microbiome research databases — is as rigorous as nutritional science currently produces.
What the AI does not do is make claims the science does not support. It does not diagnose. It does not treat. It does not override the biological individuality of the person using a MAIOME formula. What it does is compute the most scientifically defensible prebiotic composition for a given modulation goal — and do so with a precision that the best human formulator, working alone, could not achieve.
That is the appropriate use of artificial intelligence: not as a marketing term, not as a novelty, but as a genuine capability amplifier applied to a problem that genuinely requires it.
Our position: AI should earn its place in a product. In MAIOME's formulation, it earns it — by solving a computational problem that matters, with data that is real, at a level of precision that changes what a prebiotic supplement can actually be.
Why this changes what a prebiotic can be
The history of prebiotic supplementation is largely a history of single-fiber products or loosely assembled fiber blends — inulin alone, or a mixture of common fibers at round-number doses determined by cost, palatability, and the formulator's best estimate of what would be beneficial.
These products are not without value. But they are constrained by the ceiling of what human formulation can optimize. A formulator who knows that inulin feeds Bifidobacterium and that arabinoxylan feeds Ruminococcus can combine them intelligently. What a formulator cannot reliably determine is the exact ratio between those two compounds, plus quercetin, plus pectin, plus GOS, plus resveratrol — simultaneously optimized for a specific microbial target profile associated with a specific physiological goal, at milligram-level precision.
The AI pipeline removes that ceiling. It computes what human expertise approximates. And in a domain where the difference between a useful prebiotic blend and a genuinely precision-formulated one is measured in microbial ratios that have cascading effects on vitamin synthesis, mineral absorption, neurotransmitter production, gut barrier integrity, and immune calibration — the difference between approximation and precision is not academic. It is physiological.
This is what precision prebiotics actually means. Not a marketing category. A formulation methodology. And it is, to our knowledge, the most scientifically rigorous approach to prebiotic supplement design currently in existence.
Your microbiome has a blueprint. MAIOME computed it.
Every formula in the MAIOME collection was built by the same AI pipeline described above — precision-formulated for a specific physiological goal, at milligram-level accuracy no human formulator can replicate.
