EmeraldBay

# Emerald Bay Emerald Bay is a single-cell perturbation dataset of over 1.8M transcriptomic profiles spanning 52 cell lines and 91 drug treatments, including combinations. Generated using Tahoe Therapeutics's MOSAIC high-throughput platform, it comprises a curated set of anticancer agents applied at multiple doses across a MOSAIC tumor pool optimized for five-day culture. The dataset provides two readouts: a transcriptional profile at single-cell resolution and a drug-phenotype measure derived from cell-count proportions at the five-day endpoint.  ## Quickstart ```python from datasets import load_dataset # Load dataset in streaming mode ds = load_dataset("tahoebio/EmeraldBay", streaming=True, split="train") # View the first record next(ds.iter(1)) ``` Setting `streaming=True` instantiates an `IterableDataset` and prevents needing to download the full dataset first. ### Tutorials Please refer to our tutorials for examples on using the data, accessing metadata tables and converting to/from the anndata format. Please see the [Data Loading Tutorial](tutorials/loading_data.ipynb) for a walkthrough on using the data. <table> <thead> <tr> <th>Notebook</th> <th>URL</th> <th>Colab</th> </tr> </thead> <tbody> <tr> <td>Loading the dataset from huggingface, accessing metadata, mapping to anndata</td> <td> <a href="https://huggingface.co/datasets/tahoebio/EmeraldBay/blob/main/tutorials/loading_data.ipynb" target="_blank"> Link </a> </td> <td> <a href="https://colab.research.google.com/drive/1nbiNW5iJVtb2wn1iXOdvAJHhMAHqZ5BC?usp=sharing" target="_blank"> <img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open in Colab"/> </a> </td> </tr> </tbody> </table> ## Dataset Features We provide multiple tables with the dataset including the main data (raw counts) in the `expression_data` table as well as various metadata in the `gene_metadata`,`sample_metadata`,`drug_metadata`,`cell_line_metadata`, and `summary_statistics` tables. The main data can be downloaded as follows: ```python expression_data = load_dataset("tahoebio/EmeraldBay", "expression_data", split="train") ``` Per-cell transcriptomic profiles are provided (1,831,648 cells across 116 shards), with each row corresponding to one cell. The `expression_data `table has the following fields: | Field | Type | Description | |------------------------|---------------------|---------------------------------------------------------------------------------------------------------------------------------------------------| | `genes` | `sequence<int64>` | Gene token IDs for genes with non-zero expression in the cell. Aligned with `expressions`. Map to `gene_symbol`/`ensembl_id` via `gene_metadata`. | | `expressions` | `sequence<float64>` | Raw count values, aligned with `genes`. | | `drug` | `string` | Name of the treatment. `DMSO_TF` marks vehicle controls. | | `drugname_drugconc` | `string` | Compound × concentration string (e.g. `"[('Cetuximab', 0.068596, 'uM')]"`), matching the `condition` key in `summary_statistics`. | | `cell_line` | `string` | Cellosaurus ID of the cancer cell line (e.g. `CVCL_1055`). | | `sample` | `string` | Unique sample identifier (distinguishes replicate treatments). | | `BARCODE_SUB_LIB_ID` | `string` | Combination of barcode and sublibrary identifiers. Unique per cell. | ## Additional metadata ### Gene Metadata ```python gene_metadata = load_dataset("tahoebio/EmeraldBay", "gene_metadata", split="train") ``` The `gene_metadata` table maps each gene to its integer token ID used in the expression data. It extends the [Tahoe-100M gene vocabulary](https://huggingface.co/datasets/tahoebio/Tahoe-100M/viewer/gene_metadata): the first 62,710 rows preserve the Tahoe-100M token IDs verbatim, and 574 additional genes present in EmeraldBay but not Tahoe-100M are appended at the end of the vocabulary. | Column Name | Description | |---------------|-------------------------------------------------------------------------------------------------------------| | `gene_symbol` | The HGNC-approved gene symbol corresponding to each gene (e.g., *TP53*, *BRCA1*). | | `ensembl_id` | The Ensembl gene identifier (e.g., *ENSG00000000003*) based on Ensembl release 109 and genome build 38. | | `token_id` | An integer token ID used to represent each gene. This is the ID used in the `genes` field in the main data. | ### Cell Line Metadata  ```python cell_line_metadata = load_dataset("tahoebio/EmeraldBay", "cell_line_metadata", split="train") ``` Driver-mutation annotations for the 52 EmeraldBay cell lines. This is a subset of the [Tahoe-100M `cell_line_metadata`](https://huggingface.co/datasets/tahoebio/Tahoe-100M/viewer/cell_line_metadata) table filtered to the EmeraldBay cell-line panel; the schema is preserved verbatim. The table has multiple rows per cell line (one per curated driver mutation; 1–51 rows per line, mean ~9). Join on `Cell_ID_Cellosaur` to match the `cell_line` field in `expression_data` and `summary_statistics`. | Column Name | Description | |----------------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | `cell_name` | Standard name of the cancer cell line (e.g., *A549*). | | `Cell_ID_DepMap` | Unique identifier for the cell line in the DepMap project (e.g., *ACH-000681*). | | `Cell_ID_Cellosaur` | Cellosaurus accession ID (e.g., *CVCL_0023*). Join key against `cell_line` in `expression_data` and `summary_statistics`. | | `Organ` | Tissue or organ of origin for the cell line (e.g., *Lung*). | | `Driver_Gene_Symbol` | HGNC-approved symbol of a known or putative driver gene with functional alterations in this cell line (e.g., *KRAS*, *CDKN2A*). | | `Driver_VarZyg` | Zygosity of the driver variant (e.g., *Hom* for homozygous, *Het* for heterozygous). | | `Driver_VarType` | Type of genetic alteration (e.g., *Missense*, *Frameshift*, *Stopgain*, *Deletion*). | | `Driver_ProtEffect_or_CdnaEffect`| Specific protein or cDNA-level annotation of the mutation (e.g., *p.G12S*, *p.Q37*). | | `Driver_Mech_InferDM` | Inferred functional mechanism of the mutation (e.g., *LoF* for loss-of-function, *GoF* for gain-of-function). | | `Driver_GeneType_DM` | Classification of the driver gene as an *Oncogene* or *Suppressor*. | ### Drug Metadata ```python drug_metadata = load_dataset("tahoebio/EmeraldBay", "drug_metadata", split="train") ``` One row per single-drug perturbation in EmeraldBay (27 drugs; DMSO controls are excluded per the Tahoe-100M convention). Curated with Claude and validated against MedChemExpress, ClinicalTrials.gov, and PubChem. Drug-combination conditions (e.g. `Adagrasib+Cetuximab`) are not represented as rows here; join `expression_data.drug` against this table for single-drug perturbations and parse `drugname_drugconc` for combination treatments. | Column Name | Description | |--------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | `drug` | Name of the treatment. Unique key for this table. | | `targets` | Known molecular targets of the compound (gene symbol(s)). | | `mutations` | Specific target mutation(s) the compound is selective for, when applicable (e.g. `KRASG12C`). | | `moa-broad` | Broad classification of the compound's mechanism of action (typically "inhibitor/antagonist", "activator/agonist", or "unclear"). | | `moa-fine` | Specific functional MoA annotation (e.g. "RAS inhibitor", "MEK inhibitor", "Proteasome inhibitor"). | | `human-use` | "yes"/"no" — whether the compound is approved for human use. | | `clinical-trials` | "yes"/"no" — whether the compound has been evaluated in any registered clinical trials. | | `claude-notes-approval` | Contextual notes on the compound's approval status / clinical usage, generated by Claude. | | `pubchem-cid` | PubChem Compound Identifier. | | `canonical-smiles` | Canonical SMILES string representing the molecular structure (null for antibody drugs). | ### Sensitivity Readout  ```python summary_statistics = load_dataset("tahoebio/EmeraldBay", "summary_statistics", split="train") ``` Per-(cell line, condition) growth-rate summary statistics: 4,992 rows covering 52 cancer cell lines × 93 conditions (single-drug, drug-combination, and `DMSO_T0` time-zero controls). This is the raw summary table; downstream loaders typically drop `DMSO_T0`, exclude multi-drug conditions, and mean-aggregate replicates per (cell line, condition). | Column Name | Description | |----------------|--------------------------------------------------------------------------------------| | `cell_line` | Cellosaurus ID of the cancer cell line (e.g., `CVCL_0023`). | | `condition` | Compound × concentration, e.g. `[('Encorafenib', 0.1, 'uM')]`. May contain multiple tuples for combination treatments, and `DMSO_T0` marks time-zero vehicle controls. | | `growth_rate` | Scalar growth-rate response of the cell line to the treatment. |