In daily useAI-assisted analysis
LLMs and agents help us scan literature and clinical trial databases, generate code and draft reports. A computational biologist checks every number, and every conclusion traces back to its source.
APPROACH
Our Approach
Five steps from question to decision-ready delivery. AI speeds up the work, a computational biologist checks every number, and code ships with the report.
Discovery → Validation → Translation
Our work runs in three stages: find a signal in the data, test it with reproducible methods, then turn the result into something an R&D decision can rest on. The last stage matters most. What mechanism does the result point to, and does it support your next decision?
WORKFLOW
Five steps
The output of each step is the input to the next.
- 01
Define the question
We turn the R&D question into a hypothesis that data can test. "Is this target worth pursuing?" has to become something specific: in which cancer type, compared with what, measured how. If this step is off, no amount of rigor later will fix it.
- 02
Assess the data
AI-assisted scans of literature and databases, followed by an expert judgment on whether the available data can answer the question. If it can't, we tell you, and we tell you what data you would need.
- 03
Analyze and validate
Reproducible pipelines, with every number traceable to its source. Predictive models are cross-validated and externally validated.
- 04
Interpret the biology
We map results back to mechanism and clinical context. A statistically significant difference isn't the end point; we explain what it means biologically and what it changes for the decision.
- 05
Deliver for decisions
Figures and conclusions built for the meeting room and the data room.
HOW WE WORK
How we work: AI speeds us up. Experts check the work. Everything is reproducible.
Publishedmachine learning prediction models
Random Forest, SVM, Lasso and nomograms, with cross-validation and external validation. Model code and validation results ship with the report.
Exploratorygenomic foundation models
DNA language models (DNA-LLMs) combined with Hi-C / HiChIP 3D genomics, used to assess the function of non-coding variants and regulatory sequences. We offer this as an exploratory module and label it as such in reports. It is not presented as conclusive evidence.
WHAT WE DON'T DO
AI we don't do
- No black-box predictions. If we can't explain why a model works, we don't deliver it.
- No promises that "AI will find your target."
- Client data never goes into public models.
REPRODUCIBILITY
Reproducibility
- Every number traces back to its source. Each result in a report can be followed to the data and the analysis step that produced it.
- Code ships with the report. Analysis scripts and model code are part of the deliverable, so your team can check or rerun the work.
- Validation results ship with the model. Predictive models come with their cross-validation and external validation results.