Today, we announce Resolute Bio and our €10m seed round to support our work building molecular intelligence using our in house spatial proteomics platform: ResOne.
Some background:
The central dogma of molecular biology stipulates that information flows in one direction: DNA ⟶ RNA ⟶ Protein. Since 1953, scientists have been studying our DNA. More recently, the technology to study RNA has become widely available. Together, these enable the exploration of entire biological layers—genomes, transcriptomes, proteomes—giving rise to the field of "omics": the study of biological systems in their entirety.
To date, DNA and RNA have given us a many of modern biotech's meaningful innovations like DNA testing, CRISPR, mRNA vaccines, and so forth. Proteins, however, have been far more elusive.
Why is this important? It is proteins that drive function. In the quest to understand and intervene in disease, the protein is the functional layer of disease-causing biology. It is also one of the most difficult to study.
In a healthy state, things go according to plan. It logically follows that in healthy individuals, DNA, RNA and proteins are well-correlated. In a diseased state, however, our systems experience dysregulation. In disease, regulation breaks down: transcript-level and protein-level abundance frequently decouple, and the magnitude of the decoupling varies by gene, cell type, and tissue context.¹ ²
As an industry, we study biological mechanisms that are static and cannot account for post-transcriptional regulation. Together, DNA and RNA are limited surrogates that we use to predict protein expression and thereby, understand disease.
Proteins.
Proteins are much closer to disease state. Proteins are what we drug. Over 90% FDA-approved drugs target proteins - think Aspirin, Ibuprofen, JAK-inhibitors, PD-L1 inhibitors. The bulk of the biotechology industry is invested in designing proteins - one way or another.
And yet, as we have drastically expanded our modalities of therapeutic intervention from small molecules to biologics, ADCs, degraders, siRNA beyond, we have not likewise expanded our understanding of the proteins that are causingdisease.
From 20,000 genes, we estimate that millions of distinct protein entities are produced owing to cellular processes like post-translational modifications and protein isoforms. Today, only about 900 proteins have been successfully targeted for disease intervention by FDA-approved drugs. This means over 99% of the human proteome remains largely unexplored for therapeutic potential.³
The dearth of new validated targets and clinically-relevant molecular biomarkers is a fundamental lack of disease biology understanding due to inherent technology limitations.
It is also a real, and acute problem for the pharmaceutical industry. There is deep product portfolio concentration around a small set of validated targets. The clinical and preclinical markets are saturated with HER2-targeting molecules, PD-L1 inhibitors, TROP2 programs - the list goes on.
More importantly, this approach is failing patients. We are often blind to which products (drugs) should be used for which applications (patients). There is no matching system. Which non-breast cancer indications should be treated with HER2-targeting therapy? What prognostic biomarkers should we be using to qualify for ENHERTU treatment? As medicines become more precise and target more defined molecular subgroups, we need to be able to match them to patients that are likely to benefit the most - for trial success and patient benefit alike.
In 2026, when we have AI-designed anything and can sequence a genome for $100, why are these questions still so elusive?
Studying proteins is hard.
Yes, biology is complex. But it makes sense that our models of disease aren't understanding human disease biology. At best, they are trained on DNA and RNA and even then often in cell lines, not human models. In studying proteins, we have been historically limited by technology. Proteomics - the field of studying proteins - has been stubbornly resistant to Moore's law.
Mass Spectrometry
Mass Spectrometry is the most specific and powerful method of studying proteins available today. There are of course other methods- affinity and sequencing based technologies are improving steadily. They are often akin to multiplexing immunofluorescence assays - studying one protein at a time; each additional protein requires adding more reagents to the experiment. Mass spectrometry, on the other hand, is the only tested technological method available today, to get an unbiased view of the proteome.
Quietly, in the last decade, incremental innovations in this technique have compounded, resulting in over 1,000 fold improvements in throughput, depth and reliability. With mass spectrometry, we can now reliably identify and quantify tens of thousands of peptides, proteins and even proteoforms from minuscule sample volume.⁴
Even better, we are finally experiencing compounding economics: the cost to run a sample is rapidly dropping - we are now in the €10/sample territory.⁵ ⁶
This step-fold change, powered by advances in the entire ecosystem that encompasses equipment manufacturing, liquid chromatography, robotic sample preparation and deep-learning based data analysis tools, has transformed mass spectrometry based proteomics from an art into a scalable science. Further, improvements in compute and data analysis tools have made pipelines controlled and scalable across hundreds of thousands of samples. Finally, we are entering the era of clinical proteomics.
We are just beginning to fully process the implications. We can now move from the hypothesis-driven 16-protein panels to unbiased quantification of thousands of proteins. Mass spectrometry further unlocks peptide level analysis, enabling detection of low-abundance proteins, target epitope binding sites, and creating opportunities for de-novo protein discovery. Possibilities in post-translational modifications and proteoforms are also rapidly opening up, combined with better algorithms to explore the peptide and spectra space. For instance, we have seen that systematic proteogenomic re-analyses continually resurface 10⁴–10⁵ more "dark" peptides that database and antibody-based methods miss.
We fundamentally believe that proteomics is paving the way towards an entirely new domain of biological understanding and ways of classifying disease. Like anything else - proteins need context. And so, we turn to:
Digital Pathology
In parallel, advances in imaging systems, compute, and artificial intelligence has resulted in a cambrian explosion around digital pathology.
The first innings in 2017 were focused on workflow digitization - a useful and practical endeavor with ever-increasing uptake.
An extremely meaningful and perhaps somewhat unexpected result, is that we now have millions of high-resolution images of human tissue samples (whole slide images) that can be used to train large-scale pathology foundation models.
Images contain leagues more information than the human eye can process. With deep learning, we can extract that information in N dimensional space for meaningful downstream applications. With an embedding space of over 1,500 textural and morphological features, we can now identify, segment and classify individual cells and disease phenotypes from images of human tissue samples with incredible speed and accuracy. With modern compute infrastructure, it is now possible and affordable to process these giga-pixel level images at scale.
In the past years, 20+ histopathology foundation models have been released - UNI, Virchow, H-Optimus-1 and so on. These models are trained on millions of whole slide images (WSIs) and achieve state of the art performance on traditional pathology tasks like IDH mutation prediction in glioblastoma or HER2 prediction scoring. They are standardizing previously manual and error-prone tasks.
Like anything, computational pathology has limits. Fundamentally, it is an in-silico product. Recent publications start to show model limitations and diminishing returns on increased training data of images alone. For every doubling of the dataset size, we are seeing smaller and smaller percentage point improvements (think low single digit) across a host of benchmark downstream tasks. Our data further suggest there are inherent limitations to what can be predicted from a histopathology readout alone - and importantly, that molecular changes are not always reflected in morphology.
Resolute
Enter Resolute. We are a molecular intelligence company.
Our aim to redefine how we understand and treat human disease using our spatial proteomics platform: ResOne.
Resolute combines computational pathology and mass-spectrometry based proteomics. This enables an entirely new way of interrogating patient biology.
We have designed a data generation engine, specifically for high-throughput, machine-readable biological data. By standardizing the combined image and functional protein readout into an atomic unit of data and entirely controlling data generation in-house, we can flip the data quality / quantity tradeoff in biotech on its head. Our atomic unit is deliberately designed as a 1:1 match from image foundation model embeddings to the protein embedding generated on our platform.
From a single measurement, we can ask:
The translational power of these data speaks volumes. They have resulted in the successful repurposing of on-market of JAK-inhibitors to treat toxic epidermal necrolysis (TEN) in 8 patients⁷ an incoming clinical trial in TEN, another in ovarian cancer⁸, a novel target in liver fibrosis⁹ and novel early pancreatic cancer detection markers¹⁰.
At Resolute, we are unified around a single measurement unit - an atomic unit of data in biology that we call our universal data cube. This allows for large-scale, AI-ready human proteome data generation with native tissue context.
This data is also deliberately delighted to be machine readable. With these data, we can begin to build molecular intelligence - learning how morphologies are related to functional molecular profiles, how patients respond to treatment, and predict which patients will ultimately be best suited to which therapies.
Practically, we can learn how the image is related to the proteome.
Notably, our data is not limited to known human proteome sequences. With mass spectrometry as our ultimate readout, we are profiling peptides present in human tissue samples, meaning we are capturing both known and non-canonical peptides in a single measurement. This data gives rise a host of new discoveries and applications we are only beginning to understand.
Why now?
In our view, three secular trends are powering a meaningful shift where technology meets biology. First, there is increasing evidence that molecular design is beginning the march towards commoditization. Second, new therapeutic modalities are targeting increasingly specific groups of patients, defined by their molecular profiles, not only morphological ones. To be actionable, these molecular profiles need to be at the functional, proteomic layer of biology. Finally, the latest advances in artificial intelligence have enabled us to explore the data at an unprecedented pace.
Combined, these trends signal an incoming tidalwave of change. It gives rise to a host of new ways to both create and capture value. We can see some of those opportunities today – new ways of understanding and classifying disease that will inform new targets and therefore cures. We believe biomarker discovery, patient selection strategy and precision diagnostics will experience vast advances over the coming decades with data like ours. Still, there remain many applications that we have not yet fully understood.
What we do know, is that these types of data and models will power a new generation of medicine. They will help bridge the vast divide between therapeutics and diagnostics. In a not-so-distant future, we will able to programmatically identify disease in patients, programmatically design relevant cures, and thus, programmatically match the right drug to the right patient, every time. The opportunity space to build within that paradigm is vast and we are just beginning.
Thank you to our Resolute team, partners and investors for your faith in us. Onward!
¹ Buccitelli C., Selbach M. mRNAs, proteins and the emerging principles of gene expression control. Nat Rev Genet(2020). https://pubmed.ncbi.nlm.nih.gov/32709985/
² Liu Y., Beyer A., Aebersold R. On the dependency of cellular protein levels on mRNA abundance. Cell 165 (2016). https://pubmed.ncbi.nlm.nih.gov/27104977/
³ Santos R., Ursu O., Gaulton A. et al. A comprehensive map of molecular drug targets. Nat Rev Drug Discov 16 (2017). https://www.nature.com/articles/nrd.2016.230
⁴ Mund A., Coscia F., Kriston A. et al. Deep Visual Proteomics defines single-cell identity and heterogeneity. Nat Biotechnol (2022). https://www.nature.com/articles/s41587-022-01302-5
⁵ Huang S., Wang C., Lin H.J.L., Kelly R.T. The $10 proteome: low-cost, deep and quantitative proteome profiling of limited sample amounts using the Orbitrap Astral and timsTOF Ultra 2 mass spectrometers. bioRxiv (2025). https://pmc.ncbi.nlm.nih.gov/articles/PMC12324313/
⁶ Santos Gonzalez F., Hock D.H., Thorburn D.R. et al. A micro-costing study of mass-spectrometry based quantitative proteomics testing applied to the diagnostic pipeline of mitochondrial and other rare disorders. Orphanet J Rare Dis 19 (2024). https://pmc.ncbi.nlm.nih.gov/articles/PMC11605922/
⁷ Nordmann T.M., Anderton H., Hasegawa A. et al. Spatial proteomics identifies JAKi as treatment for a lethal skin disease. Nature 635 (2024). https://www.nature.com/articles/s41586-024-08061-0
⁸ Schweizer L., Kenny H.A., Krishnan R. et al. Spatial proteo-transcriptomic profiling reveals the molecular landscape of borderline ovarian tumors and their invasive progression. Cancer Cell 43 (2025). https://doi.org/10.1016/j.ccell.2025.06.004
⁹ Rosenberger F.A., Mädler S.C., Thorhauge K.H. et al. Deep Visual Proteomics maps proteotoxicity in a genetic liver disease. Nature 642 (2025). https://www.nature.com/articles/s41586-025-08885-4
¹⁰ Min J., Schweizer L., Zonderland G. et al. AI-powered Deep Visual Proteomics reveals critical molecular transitions in pancreatic cancer precursors. Cancer Discov (2026). https://doi.org/10.1158/2159-8290.CD-25-1119