Single-Cell Biology · tahoebio
Tahoe-100M
Single-cell transcriptomic atlas of over 100 million profiles across 50 cancer cell lines and 1,100 small molecules.
What Tahoe-100M is
Tahoe-100M is a large-scale single-cell perturbation dataset comprising more than 100 million transcriptomic profiles. Published by tahoebio, it measures cellular responses across 50 cancer cell lines exposed to 1,100 small-molecule compounds. The data was generated using Vevo Therapeutics' Mosaic high-throughput experimental platform to capture drug responses and cellular state transitions across varied genetic backgrounds.
The dataset addresses the demand for massive experimental corpora to build and benchmark machine learning models of cell biology. By coupling deep single-cell transcriptomics with structured chemical and biological annotations, it supports investigations in systems biology, precision medicine, and computational drug discovery.
How it was made
Experimental assays were performed by treating mixed-cell spheroids across 14 96-well plates with small-molecule perturbagens or DMSO vehicle controls. Single-cell transcriptomes were captured, generating raw count values (unique molecular identifiers) for expressed genes along with unique cell barcodes and sample identifiers.
The data is structured into an expression table containing non-zero raw counts mapped to integer gene token IDs, accompanied by five dedicated metadata tables. These provide HGNC symbols and Ensembl IDs (release 109, genome build 38), sample quality statistics (read depth, transcript counts, and mitochondrial read percentages), chemical properties (canonical SMILES, PubChem CIDs, and GPT-4o-curated targets and mechanisms validated with MedChemExpress), and cell-line driver mutation records linked to Cellosaurus and DepMap.
Key facts
- Dataset size
- Over 100 million single-cell profiles
- Cell lines
- 50 cancer cell lines
- Perturbations
- 1,100 small molecules
- Data format
- Parquet
- Assay platform
- Vevo Therapeutics Mosaic high-throughput platform
- Expression measurement
- Raw UMI transcript counts
- Reference genome build
- Ensembl release 109, genome build 38
What you can use it for
- Cellular foundation model training
- Supplies large-scale single-cell transcriptomic response data to train and evaluate machine learning models of cellular behaviour and gene regulation.
- Mechanism-of-action investigation
- Enables systematic comparison of transcriptional responses across compounds sharing broad or fine mechanism-of-action classes across multiple cancer types.
- Context-dependent response mapping
- Allows researchers to study how specific genetic backgrounds, driver gene mutations, and tissue origins modulate cellular sensitivity to drug interventions.
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/Tahoe-100M at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 1,354,800 rows of expression_data/train
BARCODE_SUB_LIB_ID
string_text- min
- 19
- median
- 19
- mean
- 19
- max
- 19
canonical_smiles
string_label- 2%
- CC12CCC3C(C1CCC2(C#C)O)CCC4=C3C=CC(=C4)OC5CCCC52%
- C1CN=C(N1)NC2=C(C=CC=C2Cl)Cl.Cl2%
cell_line_id
string_label- CVCL_05466%
- CVCL_04595%
- CVCL_04804%
drug
string_label- DMSO_TF2%
- Quinestrol2%
- Clonidine (hydrochloride)2%
expressions
list- min
- 393
- median
- 1,105
- mean
- 1269.82
- max
- 5,448
genes
list- min
- 393
- median
- 1,105
- mean
- 1269.82
- max
- 5,448
moa-fine
string_label- unclear66%
- DNA synthesis/repair inhibitor7%
- JAK/STAT inhibitor4%
plate
string_label- plate4100%
pubchem_cid
string_label- 2%
- 9046.02%
- 20179.02%
sample
string_label- smp_18082%
- smp_18752%
- smp_18692%
Strengths and limits
Good at
- Contains more than 100 million single-cell transcriptomic profiles covering 1,100 compound treatments.
- Provides comprehensive metadata linking cell-line driver mutations, chemical structures (SMILES), and molecular targets.
- Includes plate-matched vehicle controls (DMSO_TF) across all 14 experimental 96-well plates.
- Standardises biological and chemical entities using Ensembl IDs, HGNC symbols, PubChem CIDs, and Cellosaurus accessions.
Watch out for
- The first element in each gene and expression sequence represents a marker token and must be manually excluded during parsing.
- Compound mechanisms of action, molecular targets, and clinical approval indicators rely partly on automated GPT-4o curation.
- Handling the complete dataset locally requires out-of-core streaming or distributed libraries due to its scale.
Questions
Can I look inside Tahoe-100M 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 Tahoe-100M under?
Its authors declare cc0-1.0. Read the terms on the original page before reusing the data.
How are gene identifiers and counts formatted in the primary expression table?
Gene identities are stored as integer token IDs in the `genes` sequence field, aligned directly with raw counts in the `expressions` sequence field. The first entry in each sequence is a marker token that must be discarded before mapping tokens to HGNC symbols or Ensembl IDs via the `gene_metadata` table.
How are experimental controls identified?
Vehicle controls are designated as `DMSO_TF` in the `drug` field. Combining this label with the `plate` field (numbered 1 to 14) allows researchers to identify plate-matched baseline controls.
What chemical metadata is provided for the perturbagens?
The `drug_metadata` table contains canonical SMILES strings, PubChem Compound Identifiers (CIDs), predicted molecular targets, broad and fine mechanism-of-action categories, human approval statuses, and clinical trial records.
What genetic information is provided for each cell line?
The `cell_line_metadata` table provides DepMap and Cellosaurus identifiers, organ of origin, and curated driver mutations including variant zygosity, mutation type, functional consequence, and inferred gain- or loss-of-function mechanism.
How to cite it
Zhang, Jesse and Ubas, Airol A and de Borja, Richard and Svensson, Valentine and Thomas, Nicole and Thakar, Neha and Lai, Ian and Winters, Aidan and Khan, Umair and Jones, Matthew G and others. Tahoe-100M: A Giga-Scale Single-Cell Perturbation Atlas for Context-Dependent Gene Function and Cellular Modeling. bioRxiv, 2025.02.20.639398 (2025). https://doi.org/10.1101/2025.02.20.639398v1
Written from Tahoe-100M'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.