AI-Bio Convergence Institute (ABCI)
AI-Bio research that writes the hypothesis and the validation plan first
ABCI is the in-house institute of AriBnC Co., Ltd., established in April 2026. It sits inside an organisation that has run drug trials, medical device trials, human studies of health functional foods and regulatory work in Korea and abroad since 2021, and it is carrying out two prediction research programmes — strain combinations from a gut-microbiome profile, and a ‘Liver Wellness Index’ from non-invasive lifelog data. Patent applications have been filed for the technology behind both.
- ABCI established Apr 2026
- 2 AI-related patent applications filed (Aug 2025, Jun 2026)
- MOTIR industrial-innovation national R&D, 30-month agreement (Jun 2026, awarded to AriBnC)
What ABCI does
AI-Bio Convergence Institute · established April 2026 · Yongin head office
ABCI applies machine learning and AI algorithms to biological, medical and health data to find better ways to support health. It chooses among supervised learning (learning rules from labelled data), unsupervised learning (finding structure in data without labels), time-series models (reading the course of records that accumulate over time) and graph and embedding methods (turning relationships between elements into numbers), and designs new model architectures where a problem calls for one. Feature engineering (turning raw records into variables a model can read), cross-validation (splitting the data and testing on each part in turn), calibration (making a model's stated probabilities match real frequencies) and explainability (letting a person see why the model judged as it did) are standing parts of the work.
Several AI model programmes are running at once. One of them, selected as a national R&D project of the Ministry of Trade, Industry and Resources, estimates the gut-microbiome profile from lifestyle data and uses it to narrow down candidate probiotic strain combinations. Another estimates liver-related indicators from non-invasive lifelog data. The two programmes are set out in turn below.
Both programmes are at the research stage; performance and validation results will be shared, to the extent they can be published, once the research is complete.
AI GutOS — a gut-microbiome prediction platform
A MOTIR national R&D programme project under an agreement signed in June 2026; the work is currently at the research stage.
| Item | Detail |
|---|---|
| What is being researched | A model that predicts shifts in gut-microbiome composition from lifestyle data — sleep, diet, physical activity, bowel records — rather than from a stool sample. Its output is an estimated microbiome profile. |
| Programme | MOTIR's 2026 industrial-innovation talent growth programme, overseas-linked track. A 30-month agreement ending December 2028, with a total project value reported at about KRW 1.9 billion. |
| Who does the work | The research, the development and the running of the clinical work all sit with the AI-Bio Convergence Institute (ABCI) at the Yongin head office. The UNIST Research Institute (the UNIST site of our Corporate Affiliated Research Institute) supplies the basic laboratory data needed to simulate how the gut microbiome environment shifts when a particular probiotic is taken. |
| Related patent | AI-based gut microbiome profile prediction and personalised probiotics (KR 10-2025-0108178, filed August 2025). |
| What is planned | Commercialisation is planned for 2028, when the project ends; the form of the service and its target markets will be set according to the validation results and each market's regulatory requirements. |
Liver-wellness app — from non-invasive lifelog to the Liver Wellness Index
The question here: using only data collected in daily life, without blood draws or imaging, how stably can liver-related indicators be estimated — and can the result be turned into lifestyle information a user can act on?
The system was developed in June 2026, and Android and iOS apps are in preparation for Korea, Japan and Southeast Asia. The related technology was filed as a patent application in June 2026 (KR 10-2026-0099786, joint application).
- Input — non-invasive lifelog data. The items collected and the range of device integration are not yet fixed.
- Modelling task — estimate liver-related indicators from time-series lifelog data, and connect changes in the indicator to lifestyle information through defined rules.
- Output — the ‘Liver Wellness Index’ together with the items that contributed to it.
- Validation — during development the model was assessed by cross-validation against performance measures fixed in advance, and checked for reproduction in a separate group not used in training. Since development it has been re-validated on further, independently collected data, and updated accordingly.
- Scope — a wellness information service that supports lifestyle management; it is not a medical device.
Liver-wellness app — demo
What the system analyses once daily records are entered, and what it shows back: three input steps, then two result screens, played in turn. Pick a step to jump to it.
Demo screens from a build in development; the released app may differ. The values shown are demonstration inputs.
How we handle AI and data
Four standards carried through into research contracts and review documents.
Only the items the question needs, after the purpose is fixed
Each programme defines its collection scope as only the variables the research question requires, and fixes that list in the protocol.
Analysis data is handled separately from identifiers
Analysis data is kept apart from identifying information, with access rights and retention periods set per project. The EDC operating procedure used on clinical projects — granting access, keeping an audit trail of changes — is applied to research projects as well, and the detailed procedure follows the project contract and review documents.
We write down where the model does not apply
Performance is reported only against metrics fixed by the biostatistics lead before start-up, and no result is offered for data outside the model's scope. Populations the training data does not represent are written into both the report and the user-facing notes.
A person checks the wording that reaches the user
Text that reaches the user is reviewed by the people who handle MFDS and overseas regulatory work (US FDA, Health Canada, EU), so that no disease name, no wording implying treatment and no approved functional-claim wording belonging to another holder enters it. The same review rule and prohibited-wording list apply wherever a model generates the sentence.
How a research or development collaboration starts
We take proposals from principal investigators at universities and hospitals, from research leads at companies, and from clients who want an AI model built. ABCI staff at the Yongin head office review them. Please make first contact through the AI enquiries form below.
- 01
Initial discussion
- Only the topic and rough scope are exchanged. Do not send unpublished material or personally identifiable information at this stage.
- A member of staff will reply within two to five business days (KST).
- 02
NDA and data scope
- After the NDA, the data items, retention period, access rights and whether data may leave the site are fixed in writing.
- Intellectual property and how results are shared are agreed project by project; where a project connects to a national R&D programme, the terms of that agreement are checked alongside.
- 03
Writing the protocol together
- The question, endpoints, statistical analysis plan and the train/validation split rule are agreed before start-up. The statistical analysis plan and the CRF and EDC specifications are written at the Yongin head office.
- Research involving human subjects includes IRB review.
- 04
Execution and disclosure
- What AriBnC covers: study design, biostatistics, EDC build and operation, CRF design. Experiments involving strains are run at the UNIST Research Institute.
- Publication timing and the data that may be disclosed are written into the contract.
Frequently asked questions
When and where will the liver-wellness app be released?
The system was developed in June 2026 and Android and iOS apps are in preparation for Korea, Japan and Southeast Asia. We will announce the timing and order of release once they are set.
How is ABCI's research separated from the CRO business?
Client data arising from contracted clinical work is used only for the purposes of that contract and is kept apart from ABCI programme data, with separate storage and access rights.
AI enquiries
A form for contacts at universities, hospitals, companies and research institutes.