The combination of modern genetic perturbation techniques with high content screening has enabled genome-scale cell microscopy experiments that can be leveraged to construct maps of biology. These are built by processing microscopy images to produce readouts in unified and relatable representation space to capture known biological relationships and discover new ones. To further enable the scientific community to develop methods and insights from map-scale data, we have released RxRx3, the first ever public high-content screening dataset combining genome-scale CRISPR knockouts with multiple-concentration screening of small molecules (a set of FDA approved and commercially available bioactive compounds). The dataset contains 6-channel fluorescent microscopy images and associated deep learning embeddings from over 2.2 million wells that span 17,063 CRISPR knockouts and 1,674 compounds at 8 doses each. RxRx3 is one of the largest collections of cellular screening data, and as far as we know, the largest generated consistently via a common experimental protocol within a single laboratory. Our goal in releasing RxRx3 is to demonstrate the benefits of generating consistent data, enable the development of the machine learning methods on this scale of data and to foster research, methods development, and collaboration.

For more information about RxRx3 please visit RxRx.ai/rxrx3.

Please use the following format to cite this dataset as a whole:

We used the RxRx3 dataset (Fay et al. (2023). RxRx3: Phenomics Map of Biology. bioRxiv 2023.02.07.527350), available from huggingface.co/datasets/recursionpharma/rxrx3.

Compound and Gene identifiers

The RxRx3 release contains 17,063 genes, as well as 1,674 known chemical entities at 8 doses each. 16,328 of these genes are anonymized in the dataset, enabling people to explore and learn from this massive dataset while protecting Recursion’s business interests. Recursion may de-anonymize genes in this dataset in the future. If you'd like to understand more about how to get access to unblinded genes please email [email protected]. A subset of RxRx3 containing only de-anonymized wells is available on HuggingFace at RxRx3-core.

Metadata

The metadata can be found in metadata.csv in this repository. The schema of the metadata is as follows:

Attribute	Description
well_id	Experiment Name - Plate - Well (compound-004_1_AA04 or gene-088_9_Z43)
experiment_name	Experiment Name: Experiment number (compound-004 or gene-088)
plate	Plate number in the experiment (1-48)
address	Well location on the plate - "A01" to "AF48".
gene	Unblinded or anonymized gene name, or a control
treatment	Compound synonym or gene-name - guide-number (Narlaprevir or _guide_1)
SMILES	Canonical SMILES or blank for non-compounds
concentration	Compound concentration tested (in uM)
perturbation_type	CRISPR or COMPOUND
cell_type	HUVEC
well_type_label	Indicates experimental control information

The well_type_label column includes the following values:

well_type_label	Description
Query guides	CRISPR guides that target a query gene
Exon controls	Exon-targeting CRISPR guides that are used as controls
Intron controls	Intron-targeting CRISPR guides that are used as controls
Query Compounds + Intron control	Query compounds on an intron-targeting CRISPR background
CRISPR Gene Positive Controls	Control genes that are exon-targeting CRISPR guides that are used as controls, there are five genes with 6 guides each that target the exon region of the gene
Control Compounds + Intron control	Control compound on an intron-targeting CRISPR background

Metadata Example

To help understand the metadata, we have included some samples that some some of the more complex parts of the format to allow parser testing and validation

well_id,experiment_name,plate,address,gene,treatment,SMILES,concentration,perturbation_type,cell_type,well_type_label
gene-044_6_I27,gene-044,6,I27,RXRX3-65482,RXRX3-65482_guide_3,,,CRISPR,HUVEC,Query guides
compound-003_11_AD37,compound-003,11,AD37,,Phloretin,"OC1=CC=C(CCC(=O)C2=C(O)C=C(O)C=C2O)C=C1 |c:9,15,19,t:1,3,12|",0.025,COMPOUND,HUVEC,Query Compounds + Intron control
gene-065_5_O32,gene-065,5,O32,RXRX3-44488,RXRX3-44488_guide_1,,,CRISPR,HUVEC,Exon controls
gene-021_3_X40,gene-021,3,X40,RXRX3-36525,RXRX3-36525_guide_1,,,CRISPR,HUVEC,Query guides

Pre-computed deep learning-based features

To make RxRx3 useful to a variety of users, pre-computed 128-dimensional embeddings from a proprietary deep learning model (Sypetkowski, et al.) are provided for every well in the experiments in embeddings.tar.

Loading the RxRx3 image dataset

# RxRx3 is subset into 180 experiments each containing 9 - 48 plates with ~1,380 wells
from datasets import load_dataset
rxrx3 = load_dataset("recursionpharma/rxrx3", "gene-001", split="Plate1")
rxrx3 = rxrx3.sort("file_path")

## Loading embeddings and metadata for RxRx3
from huggingface_hub import hf_hub_download
import pandas as pd
import tarfile

file_path_metadata = hf_hub_download("recursionpharma/rxrx3", filename="metadata_rxrx3.csv",repo_type="dataset")
rxrx3_metadata = pd.read_csv(file_path_metadata)

# extract embedding parquet files per plate to "embeddings" folder
# format embeddings/gene-001/Plate1/embeddings.parquet
file_path_embs = hf_hub_download("recursionpharma/rxrx3", filename="embeddings.tar", repo_type="dataset")
with tarfile.open(file_path_embs, "r") as tar:
  tar.extractall()

OpenPhenom inference example

See RxRx3-OpenPhenom_inference_demo.ipynb for an example of how to load images from RxRx3 into a dataloader that concatenates the 6 channel images and computes embeddings for them using OpenPhenom, a channel-agnostic masked autoencoder trained to reconstruct microscopy images.

Changelog:

January 2023: initial release
March 2026: re-release on HuggingFace

Downloads last month: 1

Paper for recursionpharma/rxrx3

RxRx1: A Dataset for Evaluating Experimental Batch Correction Methods

Paper • 2301.05768 • Published Jan 13, 2023