01
Target first
The metabolite or pathway target is defined before any ingredient is chosen. Candidate combinations are tested in the Digital Twin simulation, and the formulation is fixed before it enters clinical study.
Approach Comparison
One Engine. Five Placebo-Controlled Trials Already Registered.
Formulation approaches differ in where design starts and in how the result is tested. We compare the first by method. For the second, we answer the questions any R&D team would ask using our own record, including the trials still running.
The Market
Where Formulation Design Starts
Comparing approaches is more useful than comparing brands. Seven approaches recur in the gut microbiome category, and the clearest difference between them is the starting point.
| Approach | Design starts from | How ingredients are selected | Live microbes |
|---|---|---|---|
| Enbiosis: target-led computational formulation | Design starts fromA metabolic target | How ingredients are selectedThe Digital Twin simulates gut microbial metabolism; AI retrosynthesis then selects a minimal set of food-grade precursors that gut microbes can convert into the target metabolites. | Live microbesNo |
| Strain-bank screening | Design starts fromA strain library | How ingredients are selectedStrains are screened for biological function, safety and manufacturability, then developed into formulations. | Live microbesYes |
| Evidence-led fibre formulation | Design starts fromIngredient evidence | How ingredients are selectedResistant starches and other fibres are combined on published ingredient evidence and a digestive or metabolic rationale. | Live microbesNo |
| Mechanism-led microbial consortia | Design starts fromA microbial function | How ingredients are selectedStrains are selected for metabolic functions such as butyrate production, using sequencing, isolation and anaerobic cultivation. | Live microbesYes |
| Gut–organ axis blends | Design starts fromAn organ axis | How ingredients are selectedBiotics and botanicals are selected, then refined using human-cell biomarker responses and preclinical testing. | Live microbesYes |
| Fermentation-guided fibre design | Design starts fromFermentation response | How ingredients are selectedFibre blends are designed from fermentation-response research to support short-chain fatty acid output across different microbiomes. | Live microbesNo |
| Strain selection with stability technology | Design starts fromImmune and barrier research | How ingredients are selectedStrains are selected using cytokine and gut-barrier research, with technology for stability and survival through digestion. | Live microbesNot stated |
Based on the methods companies in each category describe publicly, reviewed on 30 September 2026. Each row describes an approach, not every product a company makes. If a company does not describe a method publicly, that is not evidence that it lacks it.
Every approach here draws on microbiome science. Our approach differs in the order of the steps: simulation, AI retrosynthesis and precursor selection run as one workflow, from the target back to the ingredient.
Three design principles
02
No live strains
The engine designs from food-grade inputs, not live microbes. No claim depends on a strain surviving storage, transit or competition in the gut.
03
Every input has a role
Each ingredient is selected for a defined biological role in the target pathway. A longer ingredient list is not the goal.
The Technology
How the Engine Works Back from a Target
The Digital Twin & AI Engine starts from a measurable change in gut microbial metabolism and searches backwards for the ingredients and doses predicted to produce it. Each candidate is modelled in individual microbiome profiles, then run as one fixed formula across many of them.
01
Define the target
A health objective is translated into an output the model can measure: a metabolite the microbes should produce more of, or a microbe whose growth should be supported. Alongside it, limits are set on what must not change.
02
Model each microbiome
Each microbiome profile gets its own digital twin: a model of which microbes are present, which reactions they can carry out, and how they share and compete for nutrients. One ingredient can feed a microbe directly and reach another only through cross-feeding, which is why the response is modelled one profile at a time. As of October 2026, the model population covers 8,004 adult microbiome profiles and 1,007 detected species. It is a modelling cohort and does not represent every market.
03
Search backwards, then re-check
The engine uses a machine-learning model trained on these simulations to search ingredient combinations and doses quickly, staying inside the product's limits on dose, format, cost and supply. Each shortlisted formula then goes back through the full simulation, and advances only if the result holds. When no formula meets the target within those limits, the engine reports that no candidate qualified.
04
Simulate one formula across many
A product is one fixed formula used by many people. The engine runs that formula across many individual models, checking how many reach the target and how the least responsive profiles behave, rather than relying on the average alone.
Simulation narrows the design to a formula worth testing and records why each ingredient and dose was chosen, so a later change, such as a new supplier, is checked against the same specification. It is not clinical evidence. Evidence has to come from studies in people; the registered trials are set out below.
The Evidence
The Same Questions, Answered for Our Own Formulations
These are the questions R&D teams and investors ask of any formulation evidence base. Here are our answers today, including the parts still in progress.
- 01Two completed pilots
Have finished formulas been tested in people?
Two single-arm, open-label pilots are complete: the en-TEAR formula, with 20 participants (NCT07148271), and a food-bar formulation for metabolic health, with 25 participants over 12 weeks. Neither pilot had a control group.
- 02Signals to confirm
What did the pilots show?
+67%
Tear production at 16 weeks, en-TEAR pilot
- Baseline
- 10.0 mm
- Week 16
- 16.7 mm
In the en-TEAR pilot (single-arm, open-label, n=20), tear production at 16 weeks rose 67% from baseline (Schirmer test, 10.0 to 16.7 mm), eight weeks after intake had stopped. The article has been submitted for peer review and reports every secondary endpoint, including those that did not reach significance.
88%
Participants whose HbA1c at week 12 was below their baseline (22 of 25), metabolic pilot
Mean HbA1c
- Baseline
- 7.6%
- Week 12
- 6.9%
In the metabolic pilot (single-arm, open-label, n=25: 19 participants with type 2 diabetes and 6 with prediabetes), participants took an AI-designed food bar for 12 weeks. Mean HbA1c fell from 7.6% to 6.9%, and mean fasting glucose from 142.0 to 117.8 mg/dL. HbA1c was below baseline in 22 of the 25 participants (88%), and fasting glucose in 21 (84%). For both measures, p<0.001 (Wilcoxon signed-rank test). The figures come from an internal interim report, and the article has not yet been published.
Without a control group, both results are signals to confirm rather than proof.
- 03In progress
Are they being confirmed in controlled trials?
Yes. Each pilot now feeds a randomised, placebo-controlled trial: NCT07453212, with 130 participants planned, is enrolling, and NCT07622628, with 100 participants planned, is recruiting.
- 04Five registered trials
Are other formulations in clinical study?
Five randomised, placebo-controlled trials of our formulations are registered on ClinicalTrials.gov, across five health areas. The full list is in the pipeline below.
- 05Four clinical publications
What has been peer-reviewed?
Our microbiome analysis has been used in peer-reviewed clinical research on personalised nutrition: Gut Microbes (2022), Journal of Clinical Medicine (2022), American Journal of Gastroenterology (2024) and Gut Microbes (2026). These studies tested a microbiome-guided diet, not a formulation. Our first formulation results are now in peer review.
- 06Yes
Will results be reported either way?
We report results as we observe them, including endpoints that are not met.
The Pipeline
Five Randomised, Placebo-Controlled Trials
Both completed pilots now feed a randomised, placebo-controlled trial. Three more health areas have registered trials of their own: two are enrolling, and the third starts recruiting in October 2026.
CompletedMar 2026
Gut–eye axis
Pilot
Single-arm, open-label; 20 participants. Article submitted for peer review.
Completed
Primary readoutH1 2027
Gut–eye axis
Randomised trial
Quadruple-masked, placebo-controlled; 130 participants planned. Readout timing is a company estimate.
Enrolling
Primary completionMar 2027
Gut–metabolic axis
Randomised trial
Quadruple-masked, placebo-controlled, 24 weeks; 100 participants planned.
Recruiting
Primary completionMar 2027
Weight management
Randomised trial
Quadruple-masked, placebo-controlled; 80 participants planned.
Enrolling by invitation
Start plannedOct 2026
Gut and bowel health
Randomised trial
Quadruple-masked, placebo-controlled, multicentre; 162 participants planned. Primary completion expected December 2027.
Not yet recruiting
Primary completionApr 2027
Hormonal health in men
Randomised trial
Triple-masked, placebo-controlled; 112 participants planned.
Enrolling by invitation
584
participants planned across the five randomised trials.
- 130Gut–eye axis
- 100Gut–metabolic axis
- 80Weight management
- 162Gut and bowel health
- 112Hormonal health in men
Status, dates and planned enrolment come from each trial's registry entry on ClinicalTrials.gov (13 October 2026); readout timing for NCT07453212 is our own estimate. A registered trial shows that a study exists and how it is designed; it is not evidence that a formulation works.
The Confirmatory Trials Are Already Running.
For science and methodology
- See how the Digital Twin & AI Engine works
- Follow every registered trial
For partners
- License a formulation
- Source a Powered by Enbiosis active blend for your own product line
Sources
- ClinicalTrials.gov: NCT07148271, NCT07453212, NCT07622628, NCT07736547, NCT07619638, NCT07761546. Accessed 13 October 2026.
- Karakan T, et al. Gut Microbes. 2022;14(1):2138672. doi:10.1080/19490976.2022.2138672
- Arslan NÇ, et al. J Clin Med. 2022;11(22):6612. doi:10.3390/jcm11226612
- Tunali V, et al. Am J Gastroenterol. 2024;119(9). doi:10.14309/ajg.0000000000002862
- Tunali V, et al. Gut Microbes. 2026;18(1):2719125. doi:10.1080/19490976.2026.2719125
- Approach descriptions: methods described publicly by companies in each category, reviewed on 30 September 2026.
- Model population figures: Enbiosis technical white paper, Edition 2.0, October 2026 (company document).
- Metabolic pilot figures: Enbiosis Clinical & Scientific Office, Pilot Clinical Findings, internal interim report, July 2026 (company document, unpublished).
Studies 2 to 5 were sponsored by Enbiosis or include Enbiosis-affiliated authors; see each article's funding and conflict-of-interest statements.