onkommon.com
Programme R2
Digital Twins & Treatment Response Prediction
Can we predict, before treatment starts, who will progress early?
Coming soon
The problem
Treatment decisions are made against population averages. The patient in front of you is not the average, and there is currently no reliable way to say in advance which patients will progress quickly on a given plan.
Prediction models built elsewhere assume complete imaging and molecular data. In Indian practice both are frequently partial, which means models that work in a Western dataset can fail in the setting they are most needed.
The questions
- Can a patient-specific model rank treatment strategies more accurately than population guidelines?
- What is the minimum data a clinically useful model needs, in a setting where inputs are often incomplete?
- How should uncertainty be expressed so a clinician can actually use the output?
What we are contributing
We are building models for the data conditions that actually exist in India, not the ones a published paper assumes. That constraint is the research question, not a limitation to apologise for.
Because our tumour board records its reasoning in a structured, retained form, we can study whether a model would have changed a real decision, which is a far harder and more useful test than accuracy on a held-out dataset.
How the work is done
- Multi-modal machine learning across clinical, imaging and molecular inputs
- Survival and time-to-event modelling
- Mechanistic and agent-based tumour models
- Counterfactual simulation
- Uncertainty quantification, a prediction without a confidence statement is not clinically usable
What it runs on
- The institution's consented cohort
- Partner imaging under research agreement
- Public reference datasets
What it produces
- Validated predictive models
- Risk-stratification tools for board use
- Methodological publications
- Grant-funded fellowships
Why it is not just science
How this changes care here
- Risk stratification presented at the board
- The scientific basis for the molecular advisory subscription
- Earlier identification of patients who need a different plan
Funding fit
AI-in-health and computational biology grants; industry collaboration on response prediction.
What we are looking for
- Computational scientists with survival-modelling depth
- Imaging partners willing to share under research agreement
- Grant funding for compute and fellowships
Deliberately empty
Publications, funding and results
Nothing appears in this section until it exists in writing. No paper in preparation, no grant under review, no result not yet published. That rule applies with particular force to a research page.