Platform
Predict the drug. Then check it.
We are building protein-response models for drug-development research, starting with pancreatic cancer. Our future patient workflow connects a model’s shortlist with measurements from the same biopsy.
Illustrative future workflow
From biopsy to candidate drug.
Step through an invented case. Molecular-response scores are not clinical probabilities; clinical use requires outcome validation.
One biopsy, two halves.
Half A is profiled. Half B is kept to test what the model suggests.
Half A: the tumour, read.
Proteome and phosphoproteome, measured by mass spectrometry. Here, pathway activity from phosphosites.
The model ranks drugs.
Illustrative response scores and uncertainty. The top three go to experimental testing.
Half B: the top three, tested.
Each drug is applied to the other half at a few doses, and the proteome is read again.
Compare prediction and measurement.
Bar: how far each drug shut its pathway down in the patient’s tissue.
Line: what the model predicted.
A candidate for further testing.
Agreement supports further evaluation of the molecular response.
Molecular agreement does not establish tumour killing or patient benefit.
Illustration of the workflow. Not patient data; values are invented to show the steps.
The workflow
A new class of sample preparation.
Developed by our team, it holds a sample’s signalling frozen in time and reads how it answers a drug, site by site and dose by dose.
- Designed for scale.We are evaluating usable coverage, reproducibility, sample input and cost.
- Read across layers.Proteome, phosphoproteome and glycoproteome from one sample.
- Dose-resolved.Response curves at confidently measured sites across drug concentrations.
Sample · signalling in motion
The model
How the model is being built.
A compact model, pretrained on response data and adapted with paired measurements. We will compare it with RNA and simple baselines.
01
Pre-train
Learn drug- and dose-dependent changes from public data and the team’s experiments.
02
Pair
Calibrate on planned matched workflow and live-cell experiments: same contexts, drugs and doses.
03
Fine-tune
Adapt with small sets of labelled outcomes; test on independent biological contexts.
Who it’s for
Three kinds of partner.
Drug programmes
First focus: pancreatic cancer. Resistance hypotheses and combination prioritisation for RAS-pathway programmes.
Virtual-cell teams
Protein-level perturbation data for complementary model training and evaluation.
Clinical partners
A future biopsy-to-candidate workflow, contingent on clinical validation and an accredited partner lab.
See what the platform produces.
Request access, and tell us about your programme, your model or your patients.