The Biotech Betting That Cancer Drugs Fail Because Nobody Looked at the Proteins

Every year, drug developers pour billions into cancer compounds that look promising in a dish and then die in humans. The failure rate is not a rounding error. Roughly 80% of the compounds that reach clinical trials never make it, and a big share of that waste traces back to a single mismatch: the models used in the lab do not behave like the patients they are meant to represent. Yatiri Bio, a 14-person company tucked into San Diego's Sorrento Mesa biotech corridor, has built its entire business around one contrarian answer to why. Its founder thinks the field has been reading the wrong molecules.

The dominant tool of modern precision medicine is genomics - reading DNA to predict what a tumor might do. Yatiri Bio's pitch starts with an inconvenient distinction: DNA tells you what could happen. Proteins, and the chemical tags stuck onto them, tell you what is happening right now. Proteins are the machines that actually run a cell, and they are far messier and harder to measure than genes. That difficulty is exactly where Yatiri Bio has planted its flag.

01 / THE THESISRead what the cell is doing, not what it might do

The company uses high-resolution mass spectrometry - an instrument that can identify and quantify thousands of proteins in a single biological sample - and pairs it with deep learning to make sense of the flood of data. Recent gains in the technology let the team pull 8,000 or more proteins out of one sample, including the post-translational modifications (phosphorylation and the like) that flip signaling pathways on and off. Those modifications are where a lot of cancer biology hides, and where a drug's real mechanism of action often reveals itself.

Turn the raw spectra into decisions and you get Yatiri Bio's actual product: not a drug, but the evidence that stops you developing the wrong one. Its computational tools run differential expression, pathway enrichment, outlier detection and metadata associations, using machine-learning techniques like unsupervised clustering and dimensionality reduction to find structure a human analyst would miss.

02 / THE FOUNDERThirteen years at Celgene, then a detour

Yatiri Bio was co-founded in 2020 by Pilgrim Jackson and Afshin Mahmoudi. Jackson is a biophysicist by training - a Yale degree in molecular biochemistry and biophysics - who spent more than 13 years at Celgene, rising to senior principal scientist and building the protein-production and proteomics groups. That is a comfortable perch to walk away from. He left because he had watched, from the inside, how often good molecules failed for reasons nobody could see until it was too late.

The name is a tell. A yatiri is a traditional Andean healer, a reader of hidden signs who interprets what others cannot see in order to guide treatment. It is a slightly romantic name for a company that mostly runs mass specs and trains neural networks - but the metaphor is the whole thesis in one word.

03 / THE PROBLEMWhy four out of five compounds die

Yatiri Bio is blunt about the number it exists to attack. On its own materials, the company frames the core drug-development challenge this way: about 80% of the compounds that progress to clinical trials fail, largely because of poor correlation between the preclinical models and the patients they are meant to represent. If your cell line does not reflect the patient, everything you learn from it is a well-funded guess.

The fix, in Yatiri Bio's telling, is to make the models honest. Its ProteoModels are patient-derived cellular systems matched to molecularly defined cancer subtypes and backed by clinical proteome data. Instead of one generic cell line standing in for a diverse patient population, you get a portfolio of models chosen because their protein biology actually resembles a real subset of patients. Drug developers use them to optimize patient selection, test combination therapies, explore drug repurposing, and - crucially - find resistance mechanisms before a trial does.

04 / THE PRODUCTSThree tools that all start with "Proteo"

Behind the branding, the workflow is a pipeline. Raw sample goes in; a mass spec reads the proteome; the computational layer cleans, clusters and interprets it; and a partner gets back an answer that maps to a decision - which patients, which combination, which resistance path to watch. The two underlying services, Discovery Proteomics and Efficacy Models, are how that pipeline gets sold.

05 / THE MODELSelling insight, not molecules

Yatiri Bio is a B2B business. It does not carry a clinical pipeline of its own drugs to fund and defend; it partners with pharmaceutical and biotech companies and applies its platform to their programs. The customers are drug developers - concentrated in oncology today, with the company signaling expansion into immunology and neurodegeneration. Revenue comes from partnered discovery work, model access, and the collaborative data-portal engagements. It is a small, high-touch customer base, not a mass-market product, which is a sane shape for a 14-person team.

Reported funding sits at roughly $8.7 million, including a $7 million debt-financing tranche in April 2024. That is not a mega-round, and the company does not pretend otherwise. What it has instead of a giant balance sheet is an unusually deep advisory bench - names drawn from the Broad Institute, UC San Diego and Cedars-Sinai - and a habit of making its capital do more than one thing.

06 / THE 2025 RUNA grant and an acquisition in one year

Most companies this size are lucky to hit one headline milestone a year. Yatiri Bio hit two. In August 2025 it closed its first acquisition, buying San Diego neighbor NGeneBioAI - a plasma-proteomics and AI-diagnostics outfit - in an all-stock deal for an undisclosed sum. The prize was not just technology. It was infrastructure: a CLIA- and CAP-certified laboratory that lets Yatiri Bio develop and validate laboratory-developed tests. Discovery is worth far more when you can turn it into an orderable diagnostic, and most discovery-stage biotechs cannot.

Then, in November 2025, came the second: a Gates Foundation grant funding a one-year project called Proteomics Interoperability. The problem it tackles is genuinely nerdy and genuinely important. Different proteomics platforms - mass spectrometry, aptamer-based tools like SomaLogic, antibody-based tools like Olink - measure proteins in different ways and often disagree. The project trains a deep neural network to harmonize them, starting with matched tumor and adjacent-tissue samples in HPV-related cervical cancer, and promises to release a reference dataset and methods for the wider community.

07 / THE LANDSCAPEWhere it fits, and who it isn't

Yatiri Bio operates in the crowded, fast-moving band where proteomics meets AI-driven drug discovery. The platform companies - SomaLogic, Olink, Seer, Nautilus - sell instruments and reagents. The informatics teams inside big CROs sell scale. Yatiri Bio sits in a narrower slot: it is not primarily selling you a machine or a headcount, it is selling the interpretation layer plus patient-matched models, aimed squarely at the go/no-go decisions that make or break an oncology program. Buying a CLIA lab and taking a Gates grant to standardize cross-platform data are both moves that widen that slot without turning the company into a hardware vendor.

There is a version of biotech that is all hype and no substrate. Yatiri Bio is the other kind - deliberately unglamorous, science-dense, and patient. Whether proteomics gets its decade the way genomics got the last one is still an open bet. But if it does, a 14-person team in San Diego has spent five years quietly getting ready for it.

■ ■ ■