NovaGen / Services
Machine learning models for biomarker and clinical data
Diagnostic and prognostic models you can explain, with honest performance estimates instead of numbers inflated by data leakage.
Who it's for
- Biomarker teams turning a candidate panel into a classifier
- Clinical groups building risk or prognosis models from patient data
- Researchers who need a reviewer-proof check of an existing model
What you receive
AI and machine learning models
You receive: Feature selection, nested cross-validation, honest performance on held-out data, and explanations of which features drive each prediction, following TRIPOD standards.
Data we accept
Expression matrices, qPCR panels or clinical tables in Excel, CSV or SPSS, with the outcome you want to predict. We work with anonymised data only, and a confidentiality agreement can be signed before you share anything.
How a project runs
- You share your study and the data you have.
- You receive an analysis plan with the scope, timeline and a fixed price, agreed in writing before any work starts.
- Quality control comes first, and any data problems are reported as they are found.
- You receive figures, tables, documented code and a technical summary of every step, organised so you can rerun everything.
Analysis, figures and tables only; manuscript writing is not included.
Questions
- Why nested cross-validation?
- Choosing features and tuning a model on the same data used to measure it makes performance look better than it is. Nested cross-validation keeps those steps separate.
- Will I know why the model makes its predictions?
- Yes. SHAP values and model coefficients show which features drive each prediction, overall and for individual samples.
- Is a small dataset enough?
- Often, with the right model and validation. The analysis plan states what your sample size can and cannot support before work starts.
Tell us about your project
Share your study and the data you have. You'll receive an analysis plan and a fixed quote before any work starts.
Request a quote