Parse Biosciences and Nomic Bio Partner to Build 100M-Cell Cascade Immune Perturbation Dataset

The Data Ceiling in Generative AI for Drug Discovery

Machine learning models have demonstrated remarkable success in language and image processing because training engines had access to trillions of tokens scraped from the public internet. But in computational biology and drug discovery, models face an acute data bottleneck.

Predicting how human immune cells react to candidate therapeutic compounds requires multimodal, time-resolved laboratory measurements. Most historical biological datasets are fragmented: studies either measure gene expression without protein validation, test only a handful of donor cell types, or evaluate drugs at static single endpoints.

Because the human immune system relies on complex paracrine signaling cascades, training predictive foundation models requires longitudinal perturbation data that captures both transcriptomic intent and physical protein secretion. Parse Biosciences and Nomic Bio launched the Cascade initiative to build this missing empirical foundation.

Inside Cascade: 100M Single-Cell Reads and 11,000 Secretomes

Cascade is engineered as a foundational multimodal biological dataset. Rather than collecting observational patient surveys, the collaboration conducts controlled perturbation experiments at unprecedented scale:

• 100 Million Single-Cell RNA-seq Profiles: Maps transcriptional states across millions of individual immune cells, capturing gene regulation under chemical stress.
• 11,000 Quantitative Proteomic Secretomes: Measures the concentrations of over 1,000 secreted signaling proteins (cytokines, chemokines, and growth factors) directly from sample supernatants.
• 150,000 Perturbation Conditions: Exposes 14 to 18 distinct human immune cell populations to individual and combinatorial drug perturbations over multiple longitudinal intervals.

Instrument Synergy: Parse GigaLab and Nomic Omni 1000

Generating a dataset of this magnitude requires specialized instrumentation designed for high-throughput multiomics:

Parse Biosciences handles the single-cell transcriptomic pipeline through its high-capacity GigaLab facility. Parse split-pool combinatorial barcoding technology allows millions of cells to be barcoded in parallel microwell plates without requiring microfluidic droplet instruments.

Concurrently, Nomic Bio measures protein secretions using its Omni 1000 platform. The technology measures more than 1,000 secreted proteins per sample well with high sensitivity, overcoming the cross-reactivity and dynamic range bottlenecks that historically plagued multiplexed antibody assays.

Predictive Virtual Cells: Transforming Pharmaceutical Pipelines

The ultimate objective of Cascade is to train predictive virtual cell foundation models. Today, pharmaceutical sponsors spend hundreds of millions of dollars conducting exploratory wet-lab compound screening.

By training transformer architectures on Cascade multi-dimensional perturbation maps, computational biologists can simulate how human T-cells, B-cells, and macrophages respond to novel drug molecules in silico. Researchers can screen millions of virtual candidate compounds, selecting only those with verified therapeutic signaling signatures for physical synthesis.

Pharma Sponsorship: GSK, Pfizer, and Sanofi Consortium

Validating the commercial significance of the initiative, major global pharmaceutical corporations joined Cascade as foundational consortium partners, including GSK, Pfizer, and Sanofi.

These pharmaceutical leaders obtain early access to Cascade raw and processed training weights, integrating the dataset into internal oncology, immunology, and vaccine development pipelines to de-risk clinical trial asset selection.

Technical Factsheet: Cascade Dataset Parameters

The factsheet below details the primary scientific dimensions and instrumentation specifications of Cascade:

Frequently Asked Questions