How we work
Co-development and our own programmes are meant to run through the same loop: find the target, design the binder, test it in tumouroids, learn from the result.
A programme with a partner
A co-development project runs as a research collaboration. Every result comes with the validation data it rests on.
The partner brings
- An oncology target question, such as a surface target that still lacks a good binder
- The criteria a result is measured against
Oncera delivers
- Prioritised surface targets from tumouroid proteomes
- Binder candidates, scored for off-target risk and developability
- Validation data from patient-derived tumouroids, compared with healthy systems
Who owns results and inventions is set out in each collaboration agreement.
Three drug programmes in three tumour types
Our own programmes are meant to test the platform prospectively and to produce the starting points that partnerships build on.
Hepatocellular carcinoma (HCC)
A highly selective binder against a tumour-associated surface antigen. It has to tell the tumour apart from healthy liver tissue.
Pancreatic cancer (PDAC)
A starting point for a binder or an antibody-drug conjugate (ADC). It has to reach its epitope inside a stroma-rich tumour.
Cholangiocarcinoma (CCC)
A bispecific ADC concept. A second, rapidly internalising antigen is meant to carry the drug into the tumour cell.
All three programmes are preclinical. Laboratory data on the binders is still to come.
Our own harness drives our own specialist models
BEAM stands for Biomolecular Estimator of Atomic Moieties. The stack runs locally on a GPU workstation: a language model acting as the harness calls Boltz 2 to pre-sort and BEAM R2 to score the energetics, and returns an improved binder. Every round starts with a VHH sequence and ends with a better one.
1,600,000×
less compute than Flex ΔΔG at comparable ranking quality (Spearman ρ about 0.25)
An estimate based on unpublished data from the team’s project work.
VHH sequence 1
starting binder
Language-model harness
Qwen 3 80B, local. Drives our own models.
Boltz 2
open source, local, pre-sorts
BEAM R2
our own model, scores the energetics
VHH sequence 2
optimised binder
Our own biobank is patient-derived and validated on five levels
The tumouroids grow from patient tissue and are the basis the loop runs on: they supply the surface proteome for prioritisation and are the system a candidate is to be tested in. The bank has grown with funding from Bayern Innovativ, DATIpilot and the institute.
32 patient-derived models
- Hepatocellular carcinoma
- 15HCC
- Cholangiocarcinoma
- 7CCC
- Pancreatic cancer
- 10PDAC
Validated on five levels
Transcriptional
the transcriptome matches the patient tissue
Genomic
the genome matches the patient tissue
Proteomic
the surface proteome matches the patient tissue
Functional
drug response tested functionally
Freedom to operate (FtO)
rights of use settled
Figures as of October 2026.
A closed design-test-learn loop
Most pipelines run target discovery, binder design and validation as separate workstreams, and only find out at the end whether they agreed. Oncera sets them up as one loop, so that every experiment feeds the next design round. The first round with laboratory data on binders is still ahead.
- 01
Patient-derived tumouroids
Feeds the loop
- 02
Surface target prioritisation
Feeds the loop
- 03
BEAM binder generation
- 04
Binding assays
- 05
Validation in tumouroids
Each round is meant to improve the next design
- 01
Patient-derived tumouroids
Feeds the loop
MicroOrganoPits grow tumour avatars from patient material and hold the transcriptional fidelity that 2D cell lines lose.
- 02
Surface target prioritisation
Feeds the loop
Tumouroid proteomes and transcriptomes are read against hepatocyte reference organoids and public tumour-normal datasets, surfacing epitopes that abundance-driven screens miss.
- 03
BEAM binder generation
The BEAM stack generates candidate binders against the prioritised epitopes and scores them for affinity, off-target risk and developability before anything reaches a bench.
- 04
Binding assays
The most promising designs are to be synthesised and measured against the target protein by SPR and BLI. Affinity and kinetics are to decide which candidate goes into a cell system.
- 05
Validation in tumouroids
Candidates are to be tested against the same patient-derived systems they were designed for, and against healthy comparators. The comparison is meant to show whether a candidate is selective.
Tumouroids and prioritisation run once per patient line, and the loop is meant to run as often after that as it takes. The tumouroid lines are the basis it runs on: they supply the surface proteome that prioritisation reads, and they are the system a candidate is to be tested in.
The platform carries more than one modality
Oncera develops VHH binders and antibody formats for solid tumours. The binder can be the drug itself or take a payload to the tumour cell, such as a small-molecule drug, an oligonucleotide or a radionuclide.
- Antibody-drug conjugate (ADC)
- Antibody-oligonucleotide conjugate (AOC)
- Antibody-radionuclide conjugate (ARC)
For ADCs, the next project phase already tests internalisation, trafficking and in vitro toxicity (WP5.3). AOC and ARC are options for the future. Each payload has to be tested separately.
What we measure
Benchmark and laboratory figures go on this page once they have been re-run and cleared. Until then, this is what is measured, and against which baseline.
Sequence recovery of the BEAM model
against ProteinMPNN and AbMPNN on the same test set
CDR-H3 retrieval rate
against the published baseline for the same task
Tumouroid take rate for HCC and CCC
against conventional organoid culture
One piece of evidence is still open: that generated binders bind on patient-derived tumouroids and distinguish them from healthy comparators. The next project phase is meant to deliver it.
Built to run where the data is
BEAM runs on a local GPU workstation and is not run as a network service. For a co-development project it is meant to be installed at the partner, so that sequences and structures stay on the partner's premises. The page on security and compliance lists the other tools Oncera uses and where they run.
Security and compliance