What happened
Parse Biosciences and Nomic Bio announced a joint data consortium called Cascade, designed to generate a 100-million-cell perturbation that pairs single-cell transcriptomics with high-plex secretome proteomics. The initiative includes founding partners GSK, Pfizer, Sanofi, Graph Therapeutics, and Turbine, aiming to address the lack of protein secretion data in current virtual cell models.
Parse Biosciences and Nomic Bio announced on Thursday the formation of Cascade, a data consortium focused on generating a 100-million-cell perturbation . The project pairs single-cell transcriptomics from Parse’s GigaLab platform with matched, high-plex secretome proteomics from Nomic’s Omni 1000 platform. The dataset will cover more than 150,000 immune perturbation conditions, aiming to capture both gene expression changes and the resulting protein secretion profiles.
The consortium includes founding partners GSK, Pfizer, and Sanofi, along with Graph Therapeutics and Turbine. These partners bring immunology pipelines, patient-derived cell assays, and AI simulation capabilities to the project. The announcement highlights a structural gap in existing virtual cell models, which are typically trained on transcriptomics-only data that fails to capture the functional protein outputs driving immune cell communication.
Parse will use its Evercode platform to generate the RNA-seq data, leveraging a split-pool combinatorial barcoding chemistry that allows for high-throughput, low-cost profiling. Nomic will apply its nELISA technology to measure secreted proteins in cell culture supernatants, achieving high specificity and sensitivity. This combination addresses the technical challenge of pairing two fundamentally different measurement approaches on the same samples at industrial scale.
The initiative is distinct from previous open-science efforts like the Tahoe-Arc-Biohub , as Cascade is organized as a pharmaceutical consortium. This structure implies that founding partners will have preferential access to the data, with no explicit commitment to open-access release for the broader academic community. The experimental design details, including specific cell types and disease contexts, have not yet been fully disclosed.
Why it matters
Current AI virtual cell models rely on transcriptomics, which measures gene expression but not the proteins cells release to signal neighbors. This gap limits the ability of AI to predict tissue-level immune behavior, such as cytokine release or cell recruitment, which are critical for immunology drug discovery. By combining Parse’s GigaLab RNA-seq throughput with Nomic’s Omni 1000 proteomics, Cascade provides the first industrial-scale attempt to train AI on paired gene-expression and protein-secretion data. This could enable models to predict functional immune outputs rather than just transcriptional states, potentially accelerating the development of immunotherapies and reducing the need for early-stage physical experiments. However, the is not open-access, meaning academic researchers may have limited immediate access compared to the founding pharmaceutical partners.
Virtual cell models are increasingly used to predict how cells respond to drugs without physical experiments, but their accuracy in immunology is limited by the lack of protein secretion data. Transcriptomics reveals which genes are active, but not whether the resulting proteins are actually secreted to influence neighboring cells. This gap is critical because immune responses rely heavily on paracrine signaling, where secreted cytokines and chemokines coordinate tissue-level behavior.
By providing paired transcriptomic and secretomic data, Cascade allows AI models to learn the direct relationship between gene programs and functional immune outputs. This could enable more accurate predictions of drug effects on immune cell populations, such as T cell activation or macrophage recruitment, which are key targets in oncology and autoimmune disease therapies. The data may reduce the reliance on early-stage wet-lab experiments, potentially lowering costs and accelerating discovery timelines.
The involvement of major pharmaceutical companies signals that the is designed for industrial drug discovery applications rather than purely academic research. This focus may prioritize perturbation conditions relevant to current therapeutic pipelines, which could limit the dataset’s utility for broader biological questions or novel drug classes. The lack of open-access commitment may also create a divide between industry and academic researchers in the virtual cell field.
The success of Cascade will depend on whether models trained on this data demonstrate meaningful improvements in predictive performance for drug responses outside the training conditions. If the paired data leads to more accurate simulations of immune behavior, it could establish a new standard for virtual cell modeling in immunology, influencing how pharmaceutical companies approach early-stage drug screening and optimization.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
Researchers should monitor the release of Cascade’s experimental design details, including specific immune cell types and perturbation conditions, to assess the ’s generalizability. Additionally, the field will need to evaluate whether models trained on this paired data demonstrate superior predictive performance for drug responses compared to transcriptomics-only models. The timing of any broader data release or academic access agreements will also be critical for the wider virtual cell community.
The specific experimental design of Cascade, including the types of immune cells profiled and the composition of the 150,000 perturbation conditions, will determine the ’s biological relevance and generalizability. Details on whether patient-derived disease models are included will be particularly important for assessing the data’s applicability to real-world therapeutic contexts.
The field will need to evaluate the predictive performance of AI models trained on Cascade’s paired data compared to existing transcriptomics-only models. Benchmarks for drug response prediction, particularly for novel compounds or cell states outside the training set, will be critical to validate the added value of the secretome layer.
The timing and terms of any data release to the broader scientific community will be a key factor in the ’s impact. If access remains restricted to founding partners, academic researchers may continue to rely on older, open datasets, potentially slowing the adoption of multimodal virtual cell models in non-industry settings.
Future collaborations or competitive initiatives in the virtual cell space may emerge in response to Cascade, potentially accelerating the development of other multimodal datasets or AI architectures designed to integrate transcriptomic and proteomic data.