CellCarta and Imagene AI expand AI biomarker and companion-diagnostic path for pharma

CellCarta and Imagene AI said Thursday they expanded their collaboration so biopharma sponsors can validate, deploy, and scale AI-powered biomarker and companion diagnostic programs from assay development through clinical trials and regulatory readiness.

An AI read of a tissue slide only matters in drug development if it can survive validation, multi-site trials, and the companion-diagnostic grind. Pairing Imagene’s assays with CellCarta’s global lab and regulatory machine is an attempt to make that path boring enough for a drug company to actually use.

On Thursday, 1 October 2026, CellCarta and Imagene AI said they expanded a collaboration so drug companies can validate, deploy, and scale AI-powered biomarker and companion diagnostic programs. A biomarker is a measurable sign in tissue or blood that a drug team uses to tell who a medicine might help. A companion diagnostic, shortened to CDx, is a test paired with a drug and used to help decide who should get it. The release says the path runs from biomarker strategy and assay development through validation, clinical trial execution, regulatory readiness, and deployment. An assay is the lab test itself. Regulatory readiness means the package a company would take to a regulator, not a clearance already in hand. The dateline is Montreal and Miami. PR Newswire carries the release. The page stamps Oct 01, 2026, 09:15 ET, which is 9:15 a.m. Eastern. The source line is CellCarta. Those lines are the release.

What the two companies say they are combining. The line under the headline says the collaboration joins Imagene’s AI-powered assay development and biomarker prediction with CellCarta’s global companion-diagnostic and laboratory infrastructure, and that it supports active programs with leading pharmaceutical partners. The opening paragraph says the point is a coordinated path for biopharma sponsors. A sponsor, here, is the drug company paying for the program. Those lines are the release.

Who is already in the work, as the release states it. The companies say they are currently working together on programs with two leading global pharmaceutical companies. One is an AI-powered IHC companion diagnostic development program. IHC is immunohistochemistry, a stain that lights up a chosen protein in a slice of tissue so a lab can see how much of it is there and where it sits. The other applies Imagene AI’s Lung Prediction Panel in non-small cell lung cancer. Non-small cell lung cancer, shortened to NSCLC, is the most common kind of lung cancer. The release says more detail on these programs is expected in the coming months. It does not name the two drug companies. Those lines are CellCarta’s.

What Imagene’s side of a companion diagnostic is supposed to do. The release says Imagene’s OI CDx capabilities apply quantitative AI pathology to biomarker and target expression. OI is Imagene’s name for Oncology Intelligence. Quantitative, here, means a number for how much of a marker is present, not only a pathologist’s yes or no. The release lists cell and subcellular analysis, continuous and spatial scoring, and patient stratification. Subcellular means inside the cell, not only the cell as a whole. Continuous scoring is a scale, rather than a single cutoff. Spatial scoring is where in the tissue the signal sits. Stratification is sorting patients into groups a trial can use. Those lines are the release. It does not print a performance score from a named study.

What CellCarta says it adds. The release says CellCarta lets sponsors validate, deploy, and scale these AI approaches by combining fit-for-purpose analytical validation, a global clinical-trial laboratory network, companion-diagnostic development, and regulatory expertise, from development through regulatory readiness and commercialization. Fit-for-purpose means the test is checked for the job this program will ask it to do. Analytical validation is the lab work of showing the measurement holds up. Commercialization, in that sentence, is the later step of offering the test with a drug. It is the release’s description of the path. It is not a statement that a test has been approved. Those lines are CellCarta’s.

The lung panel, in the release’s own account. The collaboration also covers Imagene’s Multi-gene Lung Prediction Panel. The release says the panel predicts key NSCLC biomarkers directly from routine H&E whole-slide images within minutes. H&E is hematoxylin and eosin, the ordinary pink-and-purple stain already on most tissue slides a pathology lab cuts. A whole-slide image is a scan of that entire slide, not a single snapshot of one field. Minutes is the timing the release prints. The release calls the panel a practical case for biomarker-driven clinical trials: faster identification of patients from an H&E slide, plus the laboratory and trial infrastructure for a wider rollout. On the same page, a separate line says the Lung Prediction Panel is for Research Use Only, shortened to RUO, and not for use in diagnostic procedures. Research use only means a lab may use it in research. The release does not present it as a diagnosis. Those lines are the release.

Imagene’s product page names the eight alterations, and the October 1 release does not print that list. The page says the Lung Prediction Panel is an AI-powered research-use assay built on CanvOI, which it calls Imagene’s foundation model, and that it predicts eight genomic alterations in NSCLC from H&E whole-slide images, with results in minutes. A foundation model, here, is a large model trained on many images that a narrower assay starts from. The eight, as the page lists them, are mutations in EGFR, KRAS G12C, ERBB2, and BRAF V600E; fusions in ALK, RET, and ROS1; and MET exon 14 skipping. A mutation is a change in a gene. A fusion is two genes joined. Exon 14 skipping is a specific error in the MET gene that drops a piece of the instruction. Those are alterations targeted lung-cancer drugs already look for. The page says the results can support sample prioritization in research, including retrospective studies, tissue stewardship, and biomarker analysis. Prioritization means deciding which samples to send for a confirming molecular test first. Tissue stewardship is using a scarce biopsy carefully, so the lab does not spend the whole sample on the wrong test. Those names and uses are Imagene’s, on the product page.

How far the product page says the panel may go, and where it stops. The page says the panel may support investigational clinical-trial workflows, including sample prescreening and site-feasibility studies, and that in a prescreening workflow it can prioritize samples for confirmatory molecular testing. Confirmatory testing, the page says, is required to determine trial eligibility. A later answer on the same page is direct: the panel is for research use only, not for diagnostic procedures, and the results must not be used to diagnose disease, guide treatment, make patient-management decisions, or determine clinical-trial eligibility. The page also says this research product is not for clinical use, and it points clinical services and a laboratory-developed test to a separate OI Predict page. A laboratory-developed test is a test a single lab builds and runs itself. Imagene says it separately supports clinical laboratory work through a CAP-accredited, CLIA-certified laboratory, and that the Lung Prediction Panel is not part of those clinical offerings. CAP and CLIA are U.S. lab-quality stamps. Those limits are Imagene’s. They go further than the release’s one research-use line. The release does not say the Food and Drug Administration cleared the panel.

Two executives are quoted on the release. Michael Hreczuck, senior vice president of business development at Imagene AI, said drug companies need a practical path for turning AI-powered biomarker and image-analysis tools into assays that can support clinical development and companion diagnostic programs. He said Imagene is pleased to join forces with CellCarta, which he called a leader in clinical biomarker and companion diagnostic development, to support programs from assay development through commercialization. He said they are already applying this model across active pharma programs in both AI-powered IHC and H&E-based biomarker prediction. Christopher Ung, chief scientific business officer at CellCarta, said bringing AI-powered approaches into global drug development requires more than a strong algorithm. An algorithm, in that sentence, is the model. He said it requires the ability to validate, deploy, and scale inside the rigor of clinical development. He said the collaboration combines Imagene’s AI pathology with CellCarta’s global laboratory infrastructure, companion-diagnostic experience, and regulatory expertise, so they can validate fit-for-purpose assays, deploy them in global clinical trials, and scale programs toward regulatory readiness and commercialization. Those quotations are theirs, in the release.

Where the relationship stood before Thursday, as the release tells it. The expansion builds on Imagene’s participation in CellCarta’s Digital Pathology and AI Consortium, announced in July 2026, and extends the relationship into active biopharma development programs. The October release names that July announcement. It does not reprint it. Imagene and CellCarta will be available at the World CB and CDx Summit in Boston on October 5–8 to discuss the expanded collaboration. The announcement is dated October 1, 2026, so those dates are the meeting in that same week. The release says they will be there to discuss the work. It does not say they are giving a keynote, and it does not print a session title. Those lines are the release.

Who the about boxes say the companies are. CellCarta calls itself a leading global contract-research laboratory for the biopharmaceutical industry. A contract research organization is a lab a drug company hires instead of running every test in-house. The box says CellCarta holds CAP accreditations and CLIA certifications for specific testing methods and facilities in Canada, the United States, Belgium, Australia, and China. Specific methods and facilities is the release’s limit. It is not a claim that every test at every site carries both stamps, and it is not a clearance of the Lung Prediction Panel. The box says the lab integrates immunology, histopathology, proteomics, and genomics, along with sample management and logistics, from discovery through late-stage trials. Histopathology is tissue under a microscope. Proteomics is proteins. Genomics is genes. The site it prints is cellcarta.com. Imagene’s box says the company delivers precision oncology intelligence through multimodal foundation models and its own real-world data, for biomarker discovery, patient stratification, response prediction, and companion diagnostic development. Multimodal means more than one kind of data. Response prediction is an estimate of who may benefit from a drug. The box says the assays run from H&E-based molecular profiling to AI-IHC companion diagnostics, and that Imagene works with leading biopharma organizations from discovery through deployment. The site it prints is imagene-ai.com. The release does not print a price, a funding round, or a customer count.

The picture is Imagene’s OlPredict Lung Prediction Panel graphic. A dark field carries the OlPredict wordmark and the title Lung Prediction Panel. Eight colored nodes name NSCLC changes the panel reads from an H&E slide: EGFR, KRAS G12C, BRAF V600E, ERBB2, ALK, RET, ROS1, and MET Exon 14. A line under the title says the panel predicts key genomic alterations directly from routine H&E images within minutes, and that it is for research use only and not for diagnostic procedures. Those names and that limit are labels on the graphic. They match the research-use list on Imagene’s product page. The page’s wording for the last marker is MET exon 14 skipping. The graphic shortens that label to MET Exon 14. The graphic does not name CellCarta, it does not name a drug company, and it does not print a calendar date.

In plain terms, CellCarta and Imagene said on Thursday that they expanded a collaboration so drug companies can take AI biomarker tests from assay development through clinical trials and regulatory readiness. They say two unnamed global pharmaceutical companies are already in that work: one AI-powered IHC companion diagnostic, and one use of Imagene’s Lung Prediction Panel in non-small cell lung cancer. The release says the panel predicts key NSCLC biomarkers from a routine H&E slide within minutes, and that it is for research use only. Imagene’s own page lists eight alterations and says those results must not diagnose a patient, guide treatment, or decide who is eligible for a trial. The release does not name the two drug companies, does not print a dollar figure, and does not say the Food and Drug Administration cleared the panel.