Parse and Nomic Build 100 Million-Cell Immune Response Dataset

Cascade is intended to address a persistent data gap in biological AI: researchers have relatively limited measurements of how human immune cells respond to drugs and cytokines. Parse Biosciences and Nomic Bio are building a multimodal dataset that will pair gene activity inside cells with measurements of proteins those cells secrete. Because those signals can change over time, the project will track responses at multiple timepoints rather than treat each perturbation as a single snapshot.

That design reflects a broader challenge in modeling biology. A cell’s gene activity and its secreted proteins describe related but distinct parts of its response; either alone can leave researchers with an incomplete picture. Linking the two could give researchers more evidence for studying immune signaling and training predictive virtual cell models. The dataset’s value, though, will depend on how consistently those measurements can be connected across conditions and time.

Mapping immune responses at scale

The planned collection is large: 100 million single-cell RNA sequencing profiles and more than 11,000 proteomic secretome profiles. It will cover responses across over 150,000 individual and combined cytokine and drug perturbation conditions, measured at multiple timepoints. These figures describe the project’s intended scope; the source doesn’t specify how the final dataset will be organized or when it will be available.

The combination of many perturbations and repeated measurements is meant to let researchers compare how immune cells respond under different conditions. A drug or cytokine can affect more than one biological process, and responses may vary by cell state and timing. Capturing those differences is relevant to building models that predict what cells might do under a condition not directly tested. Still, scale alone doesn’t guarantee useful predictions: the measurements must be reliable and the experimental context clear enough for researchers to interpret them.

Pharmaceutical and biotechnology companies that have committed to the project include GSK, Pfizer, Sanofi, Graph Therapeutics and Turbine. The dataset is intended to give drug discovery teams cellular- and protein-level evidence for identifying and validating potential targets. That’s a plausible use for a resource designed around perturbations, since changing a cell’s environment can help researchers examine which biological responses follow. The announcement doesn’t describe specific targets or discoveries, so the project should be understood as building research inputs rather than reporting drug development results.

Two platforms, paired measurements

Parse will generate the RNA sequencing data at its GigaLab, a high-throughput facility capable of profiling billions of cells annually. Nomic will use its Omini 1000 platform to measure more than 1,000 secreted proteins across the samples. In practical terms, the partners are assigning different measurement tasks to specialized platforms, then seeking to bring the resulting data together.

That pairing is central to the project’s premise. RNA sequencing can show patterns of gene activity, while secreted protein measurements offer a view of signals released by cells. They aren’t interchangeable readouts, and collecting both can help researchers examine whether changes inside a cell correspond with changes in its secreted environment. (And this is the part most people overlook.) Matching measurements across samples, conditions and timepoints is a substantial part of making multimodal data useful; the announcement doesn’t explain the integration process in detail.

“Biological AI modeling has hit a data bottleneck,” said Charlie Roco, Parse Biosciences co-founder and chief technology officer. He said available data on immune responses to perturbations such as cytokines and drugs remain surprisingly limited, and that Cascade is intended to help improve the predictive power of partner AI models.

What the dataset could change

This is more significant than it looks if the resulting data can support comparisons across a broad range of immune-cell perturbations. Virtual cell models are only as informative as the conditions and measurements used to build them; a large dataset that connects gene activity with secreted proteins could give researchers more ways to test whether a model captures meaningful responses. If you're working in this space, the practical question will be whether the data make those relationships interpretable and reproducible across experiments.

There’s a clear ambition here, but the announcement describes a planned resource, not a demonstrated improvement in model performance. The stated scale is notable, yet the number of profiles doesn’t by itself establish how representative the samples are or how well the measurements will support predictions. For now, the main claim is that Cascade aims to supply richer inputs for research teams. A test of predictive value still lies ahead.

More data. Better-linked measurements. The promise is straightforward; the evidence will come from how researchers can use the completed resource.

The post 100 Million-Cell AI-Ready Dataset with Single-Cell and Proteomic Data Created by Parse and Nomic appeared first on GEN - Genetic Engineering and Biotechnology News.