From hypothesis to verifiable evidence.
Vemryx Life Sciences is a computational biomedical research lab. We use AI to read the literature, analyze public data and test hypotheses, with reproducible code and human review behind every conclusion.
The evidence ladder
Every project of ours states which step it is on. A hypothesis only moves up when the evidence supports it.
Step 1: Registered hypothesisPilot project is here
Question and hypothesis written down before any analysis.
Step 2: Reproducible analysis
Code, data and environment under version control.
Step 3: Independent replication
The finding holds in a second dataset.
Step 4: Expert review
A researcher in the field assesses methods and conclusions.
Step 5: Experimental validation
In the lab, with partner institutions.
Step 6: Clinical trial
With a sponsor, ethics approval and regulators.
- Done in our computational lab
- Only with partner institutions
What AI does here, and what it does not
Our job is to turn ideas into evidence that someone else can check. The distinction below guides everything we publish.
Generates hypotheses
Reads the literature and points to possible disease mechanisms, therapeutic targets and questions that remain unanswered.
Produces evidence
Analyzes data, tests models, attempts to reproduce published results and quantifies the uncertainty of each finding.
Does not replace the experiment
A molecule suggested by AI is not a treatment. Efficacy and safety can only be shown in biological systems and, where authorized, in people.
Research lines
Four lines, one computational lab. We open the next one only when the previous one has reproducible results.
Computational biology
Bioinformatics, genomics, therapeutic target discovery, proteins and simulations.
Foundation for the other lines
Oncology
Biomarkers, gene expression and mechanisms of treatment resistance.
Pilot project being registered
Neuroscience and psychology
Cognition, mental health, behavior and neuroimaging analysis.
Future line
Immunology and vaccines
Immune responses and computational assessment of vaccine candidates.
Future line
Pilot project
Glioblastoma is an aggressive brain cancer that is well documented in public databases. It is a good first problem because other researchers can check every step we take.
Biomarkers of therapeutic resistance in glioblastoma
- Question
- Which gene expression patterns are associated with clinical outcome in glioblastoma, and do those associations hold in an independent dataset?
- Data
- TCGA-GBM, through the NCI Genomic Data Commons (open access). The second cohort, used for replication, is set in the project record.
- Analysis
- Association between expression and clinical outcome, controlling for confounders and correcting for multiple comparisons.
- Deliverables
- A tested Python pipeline, a technical report with limitations, and every result, including the negative ones.
- Stage
- Step 1 of 6: hypothesis being registered.
The first step is to reproduce a result that has already been published. Only then do we propose something new.
What this project is not
- It is not a search for a cure in 90 days.
- It is not evidence that a new treatment exists.
- It does not use identifiable patient data.
How a question becomes a result
Five stages, always in the same order. Automation is partial: forming hypotheses, interpreting findings and publishing conclusions depend on a person’s decision.
Literature investigation
AI models search publications, compare evidence and flag gaps. Every reference is checked by DOI or PMID.
Human decision: a person chooses the hypothesis
Computational analysis
Python and R pipelines run the analyses on public data, with automated tests and version control.
Independent review
A second model and a person check the statistics, biases, reproducibility and contrary evidence. Models can be wrong together, so AI review is never the last one.
Human decision: a person interprets the findings
Researcher assessment
Specialists and partner institutions decide whether the finding justifies experimental validation.
Publication
Methods, code, limitations and verifiable results. AI use is disclosed and authorship always belongs to people.
Human decision: a person signs off on the conclusions
Vemryx Research Engine
The internal platform we are building so that any result of ours can be redone by someone else. It is in development, starting with the pilot project’s pipeline.
Data sources
Papers, studies and public datasets, with origin, version and license on record.
AI research
Agents that read the literature and propose testable hypotheses.
Requires human approval
Experiment engine
Analyses and simulations run in a controlled environment.
Validation
Statistical tests, replication and scientific review.
Requires human approval
Reproducibility
Code, data and versions kept so each result can be redone.
Publications
Technical reports and manuscripts.
Requires human approval
What every project keeps
It is what stops a platform like this from becoming a factory of bad papers.
- The scientific question and hypothesis, written before the analysis.
- The origin, version and license of each dataset.
- The code and environment needed to reproduce the result.
- Statistical tests, metrics and negative results.
- Limitations and a specialist’s assessment.
Public data, licenses respected
We work with open, documented databases. When a database holds controlled-access data, we follow its licenses, restrictions and privacy rules.
- PubMed
Papers in medicine, oncology and neuroscience.
- Open Targets
Associations between diseases and potential therapeutic targets.
- NCI Genomic Data Commons
Cancer genomics data, including TCGA.
- NCBI GEO
Gene expression studies and other functional data.
- OpenNeuro
Public neuroimaging datasets.
- ClinicalTrials.gov
Protocols and results of registered clinical trials.
For universities and research institutes
We bring engineering and method. We need people who know the biology better than we do.
We are developing a computational platform for reproducible analysis of biomedical data and biomarker identification. We are looking for scientific collaboration on methodological review, joint project development and, eventually, experimental validation.
What we offer
- Software and AI engineering dedicated to the project.
- Reproducible pipelines on public data.
- Automated literature review, checked by a person.
- Open code and auditable technical reports.
What we are looking for
- Critical review of our methods.
- Scientific guidance in biomedicine and bioinformatics.
- Joint research and development projects.
- Experimental validation when a finding justifies it.
We send a two-page technical summary covering the pilot project, the capabilities we already have and the ones we do not have yet.Request the technical summary
What we do alone and what we only do with partners
Health research has rules, and they exist for good reasons. This is the limit of what a software company should do on its own.
| Type of research | How we take part |
|---|---|
| Literature review and analysis | We do it in-house. |
| Computational studies with public data | We do it in-house, respecting licenses and access restrictions. |
| Development and evaluation of AI models | We do it in-house, with no unvalidated clinical use. |
| Experiments on cells and tissues | Only in appropriate facilities, under the supervision of partner institutions. |
| Studies with human participants | Only with prior ethics approval under the law of the country where the study takes place. |
| Clinical trials of drugs and vaccines | Only with a sponsor, a qualified team and approval from the relevant regulators. |
Sensitive data stays out of AI tools
Identifiable medical data is not sent to models without a legal basis, a contract and security measures that meet the applicable data protection law.
AI is not an author
We follow ICMJE guidance: AI use is disclosed in every manuscript and scientific responsibility lies with people.
A negative result is still a result
We publish what was not confirmed with the same care as what was.
The first 90 days
From 18 November 2026 to 16 February 2027. These are organizational goals, not promises of scientific results or signed partnerships.
Days 1 to 15
Scientific foundation
- Choose one specific research problem
- Create the repository and technical documentation
- Adopt the evidence and AI use policy
Days 16 to 30
First experiment
- Select a documented public dataset
- Reproduce a published result
- Build the pipeline with tests and versioning
Days 31 to 60
Validation and outreach
- Seek validation in independent data
- Write the technical report with its limitations
- Present the proposal to research groups
Days 61 to 90
Partnerships and funding
- Obtain a specialist’s assessment
- Prepare a funding proposal, if viable
- Define the next collaborative project
Does your group have a question that fits public data?
Tell us the question and what data exists. We reply with what we can do, how long it would take, and what we cannot do.
Writing about something else?
Methods and corrections
Ask how an analysis was done, or tell us where one is wrong.
research@vemryx.comPrivacy
Questions and requests about personal data.
privacy@vemryx.comPress
Interviews and media inquiries.
press@vemryx.comAnything else
General questions about the division.
contact@vemryx.com