mbzuai-ai4bio/BioXArena-Data-Public

BioXArena Data Public

This is the public data package for BioXArena: 9 domains, 76 tasks. Each task folder contains public/description.md, public/sample_submission.csv, the training and test inputs, and any modality-specific assets.

2026-10-14

We host the BioXArena benchmark on CodaBench. Welcome to evaluate your coding agents and climb the leaderboard.

Release 2026-10-07 (corrected data)

This release replaces all earlier versions of the data. Since the previous release, 55 of the 76 tasks were rebuilt or corrected to remove label leakage, shortcuts that let a model score well without solving the task, synthetic or misaligned labels, and input defects; several tasks were re-split (for example scaffold-, study- or site-level hold-out) and a few evaluation metrics were changed to match the redesigned tasks. The other 21 tasks are unchanged.

Scores obtained on earlier versions of a corrected task are not comparable with scores on this release. Always read each task's public/description.md, which documents the current inputs, split and metric. The corrected tasks are:

• chemical-biology (6): bace1-binding-affinity, cell-painting-perturbation, egfr-binding-affinity, gpcr-binding-multi-class, herg-binding-affinity, kinase-selectivity-multi-label
• imaging (7): drug-moa-prediction, labelfree-cell-counting, lung-nodule-malignancy, mitochondria-counting, nucleus-type-classification, skin-lesion-diagnosis, virtual-staining
• network-biology (8): gene-disease-association, go-function-multi-label, metabolic-network-kegg, pathway-membership-reactome, ppi-prediction-string, protein-complex-corum, synthetic-lethality-prediction, tf-regulatory-prediction
• perturbation-dynamics (6): cancer-drug-sensitivity, crispr-perturbation-prediction, eccite-multimodal-perturbation, gene-regulatory-network-inference, multi-timepoint-perturbation, spear-atac-perturbation
• phenotype-disease (5): alzheimers-disease-staging, autism-diagnosis, covid19-severity-classification, genotype-to-phenotype, spatial-immune-infiltration
• sequence (7): multi-tf-binding, protein-protein-interaction, remote-homology-detection, rna-protein-binding-affinity, rna-protein-binding-signal, rna-reactivity-imputation, variant-effect-pathogenicity
• single-cell (5): batch-integration, cell-type-from-expression, cross-modality-cell-matching, developmental-stage-prediction, gene-expression-denoising
• structure (6): complex-structure-evaluation, enzyme-commission-prediction, protein-binding-site-detection, protein-fold-classification, protein-protein-interface, protein-structure-prediction
• text-integrated (5): biomedical-figure-vqa, ecg-signal-qa, gene-expression-classification, medical-vqa, protein-function-matching

Each task follows the same layout:

BioXArena-Data-Public/
└── <domain>/
    └── <task>/
        └── public/
            ├── description.md
            ├── sample_submission.csv
            ├── training data
            ├── test data
            └── modality-specific assets

The exact file set varies by task. Some tasks are primarily table-based, while others also include microscopy images, molecular structures, single-cell matrices, genomic sequence files, or other multimodal resources stored under the same public/ directory.

Download From Hugging Face

This package is distributed at https://huggingface.co/datasets/mbzuai-ai4bio/BioXArena-Data-Public. There are two ways to download it.

Option A — Full package (single archive, recommended)

wget "https://huggingface.co/datasets/mbzuai-ai4bio/BioXArena-Data-Public/resolve/main/BioXArena-Data-Public.tar.gz" -O BioXArena-Data-Public.tar.gz
tar -xzf BioXArena-Data-Public.tar.gz          # extracts to BioXArena-Data-Public/<domain>/<task>/

Or with huggingface-cli:

huggingface-cli download mbzuai-ai4bio/BioXArena-Data-Public BioXArena-Data-Public.tar.gz --repo-type dataset --local-dir .
tar -xzf BioXArena-Data-Public.tar.gz

Size: 48.9 GB (sha256 cd9674d3820d15a4e62345e7defed4c47134323228abd9f58dd91fd70d458023).

Option B — One archive per domain

The same data is provided under Data-Public-by-Domain/ as 9 archives, one per domain. Each extracts to <domain>/<task>/.

huggingface-cli download mbzuai-ai4bio/BioXArena-Data-Public Data-Public-by-Domain/single-cell-10tasks-all.tar.gz --repo-type dataset --local-dir .
tar -xzf Data-Public-by-Domain/single-cell-10tasks-all.tar.gz

Download all per-domain archives at once:

huggingface-cli download mbzuai-ai4bio/BioXArena-Data-Public --repo-type dataset --include "Data-Public-by-Domain/*" --local-dir .

Data-Public-by-Domain/MANIFEST.txt lists the tasks in each archive with sha256 checksums.

Archive	Size	# Tasks	Tasks Included
Data-Public-by-Domain/chemical-biology-8tasks-all.tar.gz	23.7 MB	8	All 8 chemical-biology tasks
Data-Public-by-Domain/imaging-8tasks-all.tar.gz	24.9 GB	8	All 8 imaging tasks
Data-Public-by-Domain/network-biology-8tasks-all.tar.gz	27.3 MB	8	All 8 network-biology tasks
Data-Public-by-Domain/perturbation-dynamics-8tasks-all.tar.gz	272.4 MB	8	All 8 perturbation-dynamics tasks
Data-Public-by-Domain/phenotype-disease-8tasks-all.tar.gz	3.2 GB	8	All 8 phenotype-disease tasks
Data-Public-by-Domain/sequence-10tasks-all.tar.gz	176.8 MB	10	All 10 sequence tasks
Data-Public-by-Domain/single-cell-10tasks-all.tar.gz	4.0 GB	10	All 10 single-cell tasks
Data-Public-by-Domain/structure-8tasks-all.tar.gz	8.5 GB	8	All 8 structure tasks
Data-Public-by-Domain/text-integrated-8tasks-all.tar.gz	7.7 GB	8	All 8 text-integrated tasks

Domain Summary

Domain	# Tasks
chemical-biology	8
imaging	8
network-biology	8
perturbation-dynamics	8
phenotype-disease	8
sequence	10
single-cell	10
structure	8
text-integrated	8

Task Catalog

Chemical Biology

Task Folder	Task Title	Task Type	Evaluation Metric	Description
bace1-binding-affinity	DTI BindingDB BACE1 — Binding Affinity Regression	Regression	Pearson Correlation	Predict the binding affinity of small molecules against Beta-secretase 1 (BACE1).
cell-painting-perturbation	Cell Painting Compound Perturbation Matching	Classification	Accuracy	Predict which compound perturbation was applied to cells based on multi-channel Cell Painting morphological profiles.
cyp-inhibition-multi-label	CYP Enzyme Inhibition Multi-Label Prediction	Multi-label classification	Macro ROC-AUC	Predict whether small molecules inhibit each of five major cytochrome P450 (CYP) enzymes.
egfr-binding-affinity	DTI BindingDB EGFR — Binding Affinity Regression	Regression	Pearson Correlation	Predict the binding affinity of small molecules against Epidermal Growth Factor Receptor (EGFR).
gpcr-binding-multi-class	GPCR Ligand Family Classification	Multi-class classification	Macro F1	Predict which family of G protein-coupled receptors (GPCRs) a small molecule or peptide ligand binds, from its chemical structure.
herg-binding-affinity	DTI BindingDB hERG — Binding Affinity Regression	Regression	Pearson Correlation	Predict the binding affinity of small molecules against the hERG (human Ether-à-go-go-Related Gene) potassium channel.
kinase-selectivity-multi-label	Kinase Selectivity Multi-Label Prediction	Multi-label classification	Macro ROC-AUC	Predict the inhibition activity of small molecules against a panel of eight clinically relevant kinases.
tox21-sr-are	Tox21 SR-ARE — Oxidative Stress Toxicity Prediction	Binary classification	ROC-AUC	Predict whether a small molecule activates the Antioxidant Response Element (ARE) signaling pathway, as measured in the Tox21 stress response (SR) panel.

Imaging

Task Folder	Task Title	Task Type	Evaluation Metric	Description
amos-organ-segmentation	AMOS: Abdominal Multi-Organ Segmentation	Segmentation	Mean Dice Score	Segment 15 abdominal organs from 3D CT and MRI volumes.
drug-moa-prediction	Drug MOA Prediction	Multi-class classification	Macro F1	Predict the mechanism of action (MOA) of compounds from fluorescence microscopy images of drug-treated MCF-7 breast cancer cells.
labelfree-cell-counting	Label-Free Cell Counting	Regression	Accuracy within 10% of the true count	Predict the number of cells in label-free phase contrast microscopy images from the LIVECell dataset.
lung-nodule-malignancy	Lung Nodule Malignancy Prediction (LIDC-IDRI)	Regression	Spearman Rank Correlation	Predict the mean radiologist malignancy rating of a pulmonary nodule from the CT slices that cross it.
mitochondria-counting	Mitochondria Instance Counting in Electron Microscopy (MitoEM)	Regression	Spearman Rank Correlation	Predict the number of mitochondria instances in electron microscopy (EM) image patches from human and rat brain tissue.
nucleus-type-classification	Nucleus Type Classification	Multi-class classification	Macro F1	Predict the dominant nucleus type in H&E-stained histopathology image patches from the PanNuke dataset.
skin-lesion-diagnosis	Skin Lesion Diagnosis (HAM10000)	Multi-class classification	Accuracy	Classify dermatoscopic images of skin lesions into 7 diagnostic categories using both the image and clinical metadata.
virtual-staining	Virtual Staining — IHC Positive Ratio Prediction	Regression	Spearman Rank Correlation	Predict the immunohistochemistry (IHC) positive tissue fraction from H&E-stained histopathology images.

Network Biology

Task Folder	Task Title	Task Type	Evaluation Metric	Description
gene-disease-association	Gene-Disease Association Strength Prediction (DisGeNET)	Regression	Pearson Correlation	Predict the strength of association between a gene and a disease.
go-function-multi-label	GO Function Multi-Label Prediction	Multi-label classification	Macro ROC-AUC	Predict Gene Ontology (GO) biological process annotations for proteins.
metabolic-network-kegg	Metabolic Network Enzyme-Reaction Prediction (KEGG)	Binary classification	ROC-AUC	Predict whether an enzyme catalyzes a given biochemical reaction in the KEGG metabolic network.
pathway-membership-reactome	Pathway Membership Classification (Reactome)	Multi-class classification	Accuracy	Predict the Reactome pathway category that a protein belongs to.
ppi-prediction-string	Protein-Protein Interaction Prediction (STRING)	Binary classification	ROC-AUC	Predict whether two proteins physically or functionally interact based on their sequences and network topology features.
protein-complex-corum	Protein Complex Classification (CORUM)	Multi-class classification	Accuracy	Predict the protein complex category a protein belongs to from the CORUM database.
synthetic-lethality-prediction	Synthetic Lethality Prediction	Binary classification	ROC-AUC	Predict whether a pair of genes exhibits synthetic lethality, where simultaneous loss of both genes leads to cell death while loss of either gene alone is viable.
tf-regulatory-prediction	TF Regulatory Network Prediction (ENCODE)	Binary classification	ROC-AUC	Predict transcription factor (TF) to target gene regulatory relationships using ENCODE-derived features.

Perturbation Dynamics

Task Folder	Task Title	Task Type	Evaluation Metric	Description
cancer-drug-sensitivity	Cancer Drug Sensitivity	Regression	Mean within-drug Spearman Correlation	Predict the sensitivity (ln IC50) of cancer cell lines that are unseen in training to drug compounds.
crispr-perturbation-prediction	CRISPR Perturbation Prediction	Multi-output regression	Mean per-gene Pearson Correlation	Predict the transcriptional response to CRISPR genetic perturbations.
drug-transcriptional-response	Drug Transcriptional Response	Multi-output regression	Mean Pearson Correlation	Predict the transcriptional response of cells to drug perturbations at specific doses and in specific cell lines.
eccite-multimodal-perturbation	ECCITE-seq: Surface Protein Prediction in CRISPR-Perturbed Cells	Multi-output regression	Mean per-protein Spearman Correlation	Predict surface protein levels of CRISPR-perturbed single cells from their transcriptome.
gene-regulatory-network-inference	Gene Regulatory Network Inference	Edge prediction	AUPRC	Infer gene regulatory edges from single-cell expression data and pseudotime information.
multi-timepoint-perturbation	Multi-Timepoint Perturbation	Multi-output regression	Mean Pearson Correlation	Predict time-resolved transcriptional responses to drug perturbations.
rna-velocity-cell-transition	RNA Velocity Cell Transition	Multi-output regression	Mean Pearson Correlation	Predict unspliced RNA counts from spliced RNA counts for individual cells.
spear-atac-perturbation	Spear-ATAC: Identify the Perturbed Gene from Chromatin Accessibility	Multi-class classification	Accuracy	Identify which gene was knocked down in a single cell from its chromatin accessibility profile.

Phenotype Disease

Task Folder	Task Title	Task Type	Evaluation Metric	Description
alzheimers-disease-staging	Alzheimer's Disease Staging	Multi-class classification	Accuracy	Predict the Alzheimer's disease neuropathological change (ADNC) stage from single-nucleus gene expression profiles.
autism-diagnosis	Autism Spectrum Disorder Diagnosis (ABIDE)	Binary classification	ROC-AUC	Predict autism spectrum disorder (ASD) diagnosis from brain imaging quality metrics and phenotypic data.
breast-cancer-subtype	Breast Cancer Molecular Subtype Classification (METABRIC)	Multi-class classification	Macro F1	Predict the molecular subtype of breast cancer from clinical features and gene expression profiles.
covid19-severity-classification	COVID-19 Severity Classification	Multi-class classification	Macro F1	Predict the clinical severity of COVID-19 patients from single-cell RNA sequencing data.
diabetes-readmission	Diabetes Hospital Readmission Prediction	Multi-class classification	Macro F1	Predict whether a diabetes patient will be readmitted to the hospital within 30 days, after 30 days, or not at all.
genotype-to-phenotype	Genotype to Phenotype — Gene Expression Prediction	Regression	Pearson Correlation	Predict gene expression levels from genotype principal components and transcriptomic context.
pan-cancer-survival-prediction	Pan-Cancer Survival Prediction	Survival regression	Concordance Index	Predict patient survival risk scores from clinical and molecular features across 33 TCGA cancer types.
spatial-immune-infiltration	Spatial Immune Infiltration Prediction	Multi-output regression	Pearson Correlation	Predict the expression levels of six key immune marker genes at each spatial spot in breast cancer tissue sections.

Sequence

Task Folder	Task Title	Task Type	Evaluation Metric	Description
gene-tissue-expression	Gene Tissue Expression Prediction	Regression	Pearson Correlation	Predict gene expression levels across human tissues.
isoform-expression	RNA Isoform Expression Prediction	Multi-output regression	Mean Spearman Correlation	Predict transcript isoform expression levels across 30 human tissues.
multi-tf-binding	Multi-TF Binding Prediction	Binary classification	ROC-AUC	Predict whether a transcription factor (TF) binds to a given genomic region in a specific cell type.
protein-protein-interaction	Protein-Protein Interaction Prediction	Binary classification	ROC-AUC	Predict whether two proteins physically interact based on their amino acid sequences.
regulatory-element-detection	Regulatory Element Detection	Multi-class classification	Macro F1	Classify candidate cis-regulatory elements (cCREs) into functional categories based on their genomic coordinates.
remote-homology-detection	Remote Homology Similarity Prediction	Regression	Spearman Rank Correlation	Predict the structural similarity (TM-score) between pairs of protein domains based on their sequences.
rna-protein-binding-affinity	RNA–Protein Binding Specificity	Regression	Spearman Correlation	Predict the RNA sequence preferences of RNA-binding proteins (RBPs) that were not part of the training data.
rna-protein-binding-signal	RNA-Protein Binding Signal Prediction	Regression	Mean within-experiment Spearman Correlation	Predict the continuous eCLIP binding signal score for RNA-protein interactions.
rna-reactivity-imputation	RNA Reactivity Imputation	Multi-output regression	Mean Per-Sample Pearson Correlation	Impute missing RNA chemical reactivity values from partially observed icSHAPE in-vivo probing data.
variant-effect-pathogenicity	Variant Effect Pathogenicity Prediction	Multi-class classification	Macro F1	Predict the clinical pathogenicity of single nucleotide variants (SNVs).

Single Cell

Task Folder	Task Title	Task Type	Evaluation Metric	Description
batch-integration	Batch Integration and Cell Type Transfer	Multi-class classification	Macro F1	Transfer cell type labels to a study that was not seen during training.
cell-type-from-expression	Cell Type from Sparse Expression Profiles	Multi-class classification	Accuracy	Assign one of seven brain cell classes to single cells from extremely shallow expression profiles.
chromatin-to-expression	Chromatin to Gene Expression Prediction	Multi-output regression	Pearson Correlation	Predict gene expression (RNA) from chromatin accessibility (ATAC-seq) data at single-cell resolution.
cite-seq-protein-prediction	CITE-seq Protein Level Prediction	Multi-output regression	Pearson Correlation	Predict surface protein abundance (ADT counts) from gene expression (RNA) and protein amino acid sequences.
cross-modality-cell-matching	Cross-Modality Cell Matching	Matching	Accuracy	Match cells across two single-cell modalities: scRNA-seq (gene expression) and scATAC-seq (chromatin accessibility).
cross-modality-cell-type	Cross-Modality Cell Type Classification	Multi-class classification	Macro F1	Predict cell types from multi-modal single-cell data (CITE-seq).
developmental-stage-prediction	Developmental Stage Prediction	Multi-class classification	Accuracy	Predict the developmental stage of retinal cells after correcting for batch effects across different experimental conditions.
gene-expression-denoising	Gene Expression Denoising	Multi-output regression	Mean per-gene Pearson Correlation	Reconstruct full-depth single-cell expression counts from binomially down-sampled counts.
label-projection	Cell Type Label Projection	Multi-class classification	Accuracy	Predict cell type labels for unseen cells using a labeled reference dataset.
rna-to-protein-prediction	RNA to Protein Level Prediction	Multi-output regression	Mean Pearson Correlation	Predict surface protein (ADT) levels from RNA gene expression.

Structure

Task Folder	Task Title	Task Type	Evaluation Metric	Description
complex-structure-evaluation	Complex Structure Evaluation	Regression	Mean within-target Spearman Correlation	Score predicted structures of protein complexes against the experimental structure.
enzyme-commission-prediction	Enzyme Commission Prediction	Multi-class classification	Macro F1	Predict the primary Enzyme Commission (EC) class of a protein based on its sequence and structural features.
protein-binding-site-detection	Protein Binding Site Detection	Binary classification	AUPRC	Predict whether a protein chain has high binding-site density.
protein-fold-classification	Protein Fold Classification	Multi-class classification	Accuracy	Predict the structural fold class of a protein domain.
protein-ligand-binding-affinity	Protein-Ligand Binding Affinity	Regression	Pearson Correlation	Predict the binding affinity (pK value) of protein-ligand complexes.
protein-protein-interface	Protein–Protein Interface Residue Prediction	Residue-level binary classification	Mean per-chain AUPRC	Given one protein chain and the sequence of its binding partner, predict which of its residues form the interface.
protein-stability-change	Protein Stability Change	Regression	Spearman Correlation	Predict the change in thermodynamic stability (ddG) caused by single amino acid mutations in proteins.
protein-structure-prediction	Protein 3D Structure Prediction	Structure prediction	TM-score	Predict the 3D structure of a protein from its amino acid sequence.

Text Integrated

Task Folder	Task Title	Task Type	Evaluation Metric	Description
biomedical-figure-vqa	Biomedical Figure Visual Question Answering (PMC-VQA)	Multiple-choice VQA	Accuracy	Answer multiple-choice questions about biomedical figures extracted from PubMed Central (PMC) scientific articles.
dna-enzyme-function	DNA Enzyme Function Classification (BioTalk)	Multi-class classification	Accuracy	Predict the Enzyme Commission (EC) class for a gene given its DNA nucleotide sequence and contextual information.
ecg-signal-qa	ECG Signal Question Answering (ECG-QA)	Open-ended QA	Accuracy	Answer clinical questions about 12-lead electrocardiogram (ECG) recordings.
gene-expression-classification	Gene Expression – Text Matching (CellWhisperer)	Binary classification	ROC-AUC	Decide whether a short text description matches a gene expression profile.
medical-vqa	Medical Visual Question Answering (SLAKE)	Open-ended VQA	Accuracy	Answer open-ended clinical questions about medical radiology images.
molecule-qa	Molecule Question Answering (MoleculeQA)	Multiple-choice QA	Accuracy	Answer multiple-choice questions about molecules given their SMILES (Simplified Molecular-Input Line-Entry System) representation.
pathology-vqa	Pathology Visual Question Answering (PathVQA)	Open-ended VQA	Accuracy	Answer questions about pathology images.
protein-function-matching	Protein-Function Text Matching (SwissProtCLAP)	Binary classification	ROC-AUC	Decide whether a free-text functional annotation describes a given protein.

How To Use

1. Choose a task under a domain.
2. Open that task's public/ directory.
3. Read public/description.md first.
4. Load the task-specific public inputs from the same public/ directory.
5. Generate predictions following public/sample_submission.csv.

Notes

• Public artifacts are heterogeneous across tasks. Depending on the task, public/ may contain tables, images, compressed arrays, single-cell objects, sequence files, or other modality-specific assets.
• The exact input and output expectations are task-specific, so description.md is the authoritative entry point for each task.
• Row order and id values carry no information about the labels; do not rely on them.

License

The BioXArena data is released under CC BY-NC 4.0, for academic and other non-commercial use. Each task is derived from public sources (cited in its public/description.md), and use of a task's data is also subject to the license and terms of use of its original source. The code (runners and graders) is available at https://github.com/mbzuai-ai4bio/BioXArena under the MIT License.