Veritope uses machine learning to identify which peptides a tumor presents on its surface, and which of those a T cell can recognize. We start with naturally occurring cancers in dogs, with the aim of translating what we learn to human oncology.
Most computational pipelines rank tumor peptides by predicted binding to MHC molecules. Binding is necessary for presentation but not sufficient. Many predicted binders are never detected on the cell surface.
A presented peptide is a useful target only if a T cell recognizes it and does not also recognize a similar peptide on healthy tissue. Each of these conditions needs its own evidence.
Presentation, recognition and design rely on different data. We model them separately and evaluate each against published methods on data held out from training.
Which peptides does a given tumor present?
Models trained on immunopeptidomics, the mass-spectrometry measurement of peptides eluted from MHC molecules, rather than on predicted binding alone. Public data for the canine MHC (DLA) are far sparser than for human HLA, and that gap is part of the work.
Which presented peptides can a T cell recognize, and only on the tumor?
Models of T-cell receptor recognition of peptide–MHC complexes, evaluated on receptors and peptides absent from training, with near-identical self peptides as negative controls.
Where no natural T-cell response exists, can one be engineered?
Computational design of binding proteins against a verified target, screened against look-alike peptides from healthy tissue before any laboratory work.
A hypothesis is never presented as a result. On this site and in our notes, each claim falls into one of these categories.
Published, reproducible findings. We cite the source.
Hypotheses we have not yet tested. Each states the result that would refute it.
Experiments with the primary metric, the baseline to beat and the failure threshold written down before we see results.
Pet dogs develop cancer spontaneously, with intact immune systems, in the same homes as people. Several canine cancers, including osteosarcoma, lymphoma and melanoma, resemble their human counterparts in biology and clinical course, and they progress faster, so outcomes are observed sooner.
Starting here is also the faster path to impact. Canine studies read out in months rather than years; the findings stand on their own in veterinary oncology, reaching families seeking options for their dogs, while the same methods build the evidence to move responsibly toward human disease.
No animal model predicts human outcomes perfectly. Translation to human cancer is our long-term aim, and we treat it as a hypothesis to test. NCI overview of comparative oncology ↗
Our first milestones use public datasets only. We publish the data sources we rely on, our hypotheses before we test them, and results that fail as well as those that succeed.