Accelerating drug-efficacy prediction for pancreatic cancer
The protein layer of the virtual cell.
Drug-response data and AI, built to guide targeted-therapy research through protein response.
Developed by researchers atJohns Hopkins Whiting School of EngineeringJohns Hopkins School of Medicine
Why pancreatic cancer
Deadly, and RAS is now druggable.
In 2026 the FDA approved a RAS inhibitor for pancreatic cancer. In its trial, 30% of patients responded, against 11% on standard chemotherapy. Knowing in advance who will respond is the open question.
people in the US expected to be diagnosed with pancreatic cancer in 2026
five-year relative survival when diagnosed after spread to distant organs, as about half are
of pancreatic adenocarcinomas carry a KRAS mutation
industry-sponsored drug trials recruiting pancreatic cancer patients worldwide, including multi-tumour trials
Sources: NCI SEER (2026 estimates, with survival for 2016–2022 diagnoses), TCGA, Cancer Cell 2017, FDA, August 2026 and a ClinicalTrials.gov query of recruiting, industry-sponsored interventional drug trials (1 October 2026).
The patient path · future application
One biopsy, two halves.
Our intended workflow for patients begins with pancreatic cancer. One half of a biopsy profiles the tumour, and the other half tests the model’s shortlist. Clinical use will require validation against patient outcomes.
Split
A biopsy is divided into two halves.
Profile
One half is read by proteomics.
Predict
The model ranks drugs by predicted response.
Test
The top candidates are tested on the other half.
Compare
Where prediction and measurement agree: a candidate for further testing.
The platform
Built from the sample up.
A new class of sample preparation.
Our sample-preparation and analytical methods are designed to turn signalling, frozen in time, into informative training data.
Three layers from one sample.
Proteome, phosphoproteome and glycoproteome, site by site and dose by dose.
A model built to guide the next experiment.
Our first focus is pancreatic cancer: prioritising resistance hypotheses and combinations across RAS-pathway drugs. The aim is a reusable model, checked against measured outcomes.
Illustrative molecular-response scores. The values are invented and are not clinical probabilities.
The data
Dose by dose, site by site.
We are combining public dose–response datasets with our own into one training set for the model, with proteins, phosphorylation and N-glycosylation measured dose by dose.
drug dose–response curves from in-house and public sources, being harmonised into one training set
drug × cell-line dose series from in-house and public sources
layers from each sample: proteome, phospho and N-glyco
Publications
Built on the team’s published work.
Dose-Dependent Inhibition of Protein Glycosylation Reveals Crosstalk between Glycosylation and Phosphorylation in EGF Signaling Pathways
Analytical Chemistry · 2026Characterizing the Effects of Protein Glycosylation Perturbation on Phosphorylation Signaling
Research Square · 2026Attenuation of EGFR-Mediated Cellular Signaling by Sialidase-Mediated Desialylation
Working on a drug, a model or a patient cohort?
Tell us about your drug programme, the decision you need to make and the data available.