Single-Cell Perturbation · tahoebio
EmeraldBay
Single-cell perturbation dataset pairing transcriptomics and drug sensitivity across 52 cell lines and 91 treatments.
What EmeraldBay is
Emerald Bay is a single-cell perturbation dataset published by Tahoe Therapeutics, comprising more than 1.8 million transcriptomic profiles. It captures cellular responses across 52 cancer cell lines exposed to 91 drug treatments, including single agents at multiple doses and combination regimens.
Generated using Tahoe's MOSAIC high-throughput platform, the resource is built upon a pooled tumor cell culture optimized for a five-day assay. It delivers dual readouts: transcriptional measurements at single-cell resolution and a drug-phenotype metric derived from cell-count proportions at the five-day endpoint.
The dataset includes raw count matrices alongside curated metadata detailing driver mutations across cell lines, mechanism-of-action classifications, and chemical structures. This allows researchers to study the relationships between genetic alterations, transcriptional regulation, and cancer cell viability.
How it was made
Emerald Bay was produced on Tahoe's MOSAIC high-throughput platform by culturing pools of 52 cancer cell lines for five days in the presence of curated anticancer compounds, drug combinations, or vehicle controls (DMSO). Transcriptomic profiles were captured at single-cell resolution, recording raw counts for non-zero genes matched to an integer token vocabulary.
To quantify drug efficacy, relative cell counts were measured at the five-day culture endpoint to calculate scalar growth-rate values for each cell line and perturbation condition. The accompanying metadata tables link tokenized genes to HGNC and Ensembl identifiers, cell lines to DepMap accessions and driver mutations, and single drugs to molecular targets and SMILES representations.
Key facts
- Dataset size
- 1,831,648 single cells across 116 shards
- Cell lines
- 52 cancer cell lines
- Treatments
- 91 drug treatments (including combinations) and vehicle controls
- Single compounds catalogued
- 27 anticancer drugs
- Culture duration
- 5 days
- Assay platform
- Tahoe MOSAIC high-throughput platform
- Primary readouts
- Single-cell RNA expression (raw counts) and scalar growth rate
- Summary statistics records
- 4,992 condition-level rows across 52 cell lines and 93 conditions
- Data format
- Parquet
What you can use it for
- Predicting Perturbation Transcriptomics
- Training and benchmarking computational models that forecast single-cell gene expression changes following anticancer drug treatment.
- Drug Sensitivity and Viability Modelling
- Evaluating how baseline expression and genetic backgrounds influence phenotypic drug sensitivity and scalar growth-rate responses.
- Investigating Drug Combinations
- Analysing differential expression profiles and growth outcomes induced by multi-drug treatments compared to single-agent exposures.
- Linking Oncogenic Drivers to Treatment Response
- Studying the functional impact of specific driver mutations, zygosity, and mechanism types on therapeutic susceptibility across cell lines.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from tahoebio/EmeraldBay at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 189,488 rows of expression_data/train
BARCODE_SUB_LIB_ID
string_text- min
- 16
- median
- 16
- mean
- 16
- max
- 16
cell_line
string_label- CVCL_054617%
- CVCL_10986%
- CVCL_04596%
drug
string_label- DMSO_TF4%
- Tucatinib+Trastuzumab4%
- GDC-60364%
drugname_drugconc
string_label- [('DMSO_TF', 0.0, 'uM')]4%
- [('Sotorasib', 0.005, 'uM'), ('LY3200882', 0.5, 'uM')]2%
- [('Cetuximab', 0.068596, 'uM'), ('Adagrasib', 0.005, 'uM')]2%
expressions
list- min
- 263
- median
- 1,325
- mean
- 1672.98
- max
- 6,846
genes
list- min
- 263
- median
- 1,325
- mean
- 1672.98
- max
- 6,846
sample
string_label- smp_8502%
- smp_8422%
- smp_8042%
Strengths and limits
Good at
- Pairs single-cell gene expression readouts with matched phenotypic growth-rate measurements across identical cell line and treatment pairs.
- Profiles a large cohort of over 1.8 million cells covering 52 distinct cell lines and 91 treatment regimens.
- Contains extensive annotations, including driver mutations, drug molecular targets, and chemical SMILES strings.
- Supports streaming mode for low-memory access across 116 Parquet shards without requiring complete local downloads.
Watch out for
- The structured drug metadata table covers only the 27 single compounds, leaving combination regimens to be parsed from condition strings.
- Readouts are limited to a single five-day culture endpoint without interim temporal sampling.
- Perturbation testing is constrained to the 52 cancer cell lines included in the MOSAIC pool.
Questions
Can I look inside EmeraldBay without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is EmeraldBay under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
What readouts are included in the dataset?
Emerald Bay provides two paired readouts: per-cell raw transcriptomic count sequences and a phenotypic growth-rate readout derived from cell proportions at the end of the five-day culture.
How are genes identified in the expression matrices?
Genes are represented by integer token IDs aligned to raw expression counts. These map to HGNC symbols and Ensembl IDs via the gene metadata table, which incorporates the 62,710 Tahoe-100M tokens plus 574 appended genes.
Where can annotations for drug combinations be found?
Combination details are stored as compound-concentration tuples within the condition and treatment columns of the expression and summary tables, as the drug metadata table catalogues only the 27 single-agent perturbations.
How are cancer cell lines cross-referenced with public repositories?
The cell line metadata table provides both Cellosaurus accession numbers and DepMap identifiers for all 52 profiled lines, alongside organ of origin and curated driver mutation profiles.
Written from EmeraldBay's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.