Therna releases Chronos data while keeping its RNA models private

Why it matters

Therna is exposing wet-lab data outsiders can test while withholding its experimental protocols, RNA foundation models and a separate immune-cell study.

Nazli Azimi and Hani Goodarzi launched Chronos on September 15th, releasing a sample of the experimental data that feeds Therna Biosciences' AI system for designing RNA medicines.

The San Francisco biotech says Chronos measures tens of thousands of synthetic genes and messenger RNAs across 50 human cell lines over time, producing millions of functional measurements in one experiment. Therna made two modules, Penta-47x27K and Tria-47x28K, available through a public collection on Hugging Face, with an interactive Lightning Studio and a bioRxiv preprint describing quality controls and early modeling work.

Therna's launch is a calculated reveal. Researchers get enough material to test sequence-to-function models and examine how RNA behavior changes by cell type. Therna retains the experimental protocols, its RNA foundation models and a separate Chronos study conducted in human immune cells.

That boundary defines the business strategy. The public study gives outside researchers a way to assess Therna's premise, while the components most directly tied to drug programs and partnerships remain inside Therna.

The product is the experiment

RNA-Logix is Therna's name for a system that combines foundation models, generative AI and laboratory data to design messenger RNA or identify sites that can be targeted with antisense oligonucleotides and small interfering RNA. Chronos supplies the measurements that connect a proposed sequence with what it actually does inside a cell.

Therna says conventional experiments often study one RNA in one cell type at one point in time. Chronos pools 50 human cell lines supplied by Tahoe Therapeutics and follows thousands of synthetic sequences across those different cellular settings.

The distinction matters because an RNA sequence does not operate independently from the cell reading it. A sequence that produces a desired level of expression in one cell type may behave differently in another. Goodarzi said in the launch announcement Chronos is designed to capture that variation so a model can learn how sequence, cellular context and time combine to determine function.

The two released modules form a synthetic gene atlas for learning regulatory patterns across cellular contexts, according to the preprint. Researchers can use the data to train models, compare predictions and test rules related to expression, stability and cell-specific behavior.

Processing the resulting single-cell data is a separate computational problem. Therna reports an approximately 200x speedup using rapids-single-cell, an open-source package accelerated by NVIDIA CUDA-X. Therna has not supplied an independent benchmark for that figure, so the number should be read as Therna's measurement of its own workflow.

Azimi is building after a biotech exit

Chronos reflects the backgrounds of Therna's two principal scientific founders. Azimi trained in pharmacy and immunology, earning a Pharm.D. and Ph.D. from the University of Tehran before completing postdoctoral work at the National Cancer Institute. She previously founded and led Dermaheal and Bioniz Therapeutics, both focused on immunological disease.

Equillium acquired Bioniz in 2022. An SEC filing shows that the transaction included stock consideration and potential payments of up to $307.5 million tied to regulatory and commercial milestones. The structure gave Azimi experience with the full biotechnology cycle: building a therapeutic program, taking it into clinical development and negotiating an exit whose ultimate value depended on later drug progress.

At Therna, Azimi has returned to immunology with a broader design system. She has framed the founding thesis in linguistic terms: "RNA is a language, and a company able to understand it can write medicines in it." The statement appeared in the launch announcement. Chronos is the data-gathering machinery intended to turn that metaphor into measurable design rules.

Goodarzi brings the computational and academic side. He earned a Ph.D. in computational biology and genomics from Princeton and has led a UCSF research group since 2016, working across machine learning, RNA regulation and cancer biology. He is also a core investigator at the Arc Institute, where his lab develops AI models for functional genomics and studies RNA programs involved in cancer progression.

Therna began operating publicly in 2025 after stealth work and raised a $10 million seed round co-led by AIX Ventures, Pear VC and Fusion Fund. That financing backed a lab-in-the-loop thesis: Therna would generate biological measurements, use them to train models, test the models' designs and feed the results into another cycle.

Chronos is the clearest view Therna has provided of that loop's experimental side.

A public sample with a private center

AI drug developers increasingly argue that proprietary biological data will separate useful systems from models trained mostly on public information. Asimov has described its RNA Edge system, combining computational design with laboratory validation. Alnylam signed an AI collaboration with Inceptive valued at up to $2 billion to apply foundation models to RNA interference drug discovery.

Those deals put commercial pressure on smaller developers to demonstrate that their models are grounded in biological measurements that competitors cannot easily reproduce. Therna is responding by publishing a portion of its data operation rather than presenting another model benchmark in isolation.

The release still leaves the central scientific test ahead. The preprint describes the dataset and preliminary sequence-to-function modeling, but Chronos has yet to establish cross-laboratory reproducibility or show that its measurements improve the clinical performance of an RNA medicine. The public material supports Therna's claim that it can produce a large, varied dataset. It does not yet prove that RNA-Logix can turn that dataset into a successful drug.

Azimi and Goodarzi have nevertheless chosen a useful form of transparency. Other researchers can inspect the released study, build against it and publish competing results. Therna is withholding its experimental protocols, RNA foundation models and separate immune-cell study. If outside work validates the public dataset, Therna gains credibility without disclosing those private assets.