跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
其他技能

datasets-loading

OmicVerse built-in datasets: pbmc3k, pancreas, dentategyrus, zebrafish, immune, spatial, multiome, plus create_mock_dataset() and predefined_signatures GMT gene sets.

omicverse · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add omicverse/omicclaw --skill datasets-loading

技能说明


name: datasets-loading title: OmicVerse built-in datasets and mock data description: "OmicVerse built-in datasets: pbmc3k, pancreas, dentategyrus, zebrafish, immune, spatial, multiome, plus create_mock_dataset() and predefined_signatures GMT gene sets."

OmicVerse Built-in Datasets

ov.datasets provides 30+ ready-to-use datasets with automatic download, caching, and fallback to mock data. Use these instead of manually downloading files or relying on scanpy.datasets.

When to Use This Module

  • Tutorials/demos: Load standard benchmarks (PBMC3k, Paul15, dentate gyrus) with one function call
  • Testing pipelines: Use create_mock_dataset() to generate synthetic data without downloads
  • Gene set analysis: Use predefined_signatures for curated GMT gene sets (cell cycle, gender, mitochondrial, tissue-specific)
  • Velocity workflows: Load pre-formatted datasets with spliced/unspliced layers

Dataset Catalog

Single-Cell

FunctionCellsGenesDescription
ov.datasets.pbmc3k()2,70032,73810x PBMC3k (raw or processed)
ov.datasets.pbmc8k()~8,000—10x PBMC 8k
ov.datasets.paul15()2,7303,451Myeloid progenitors
ov.datasets.krumsiek11()64011Myeloid differentiation simulation
ov.datasets.bone_marrow()5,78027,876Bone marrow hematopoietic
ov.datasets.hematopoiesis()——Processed hematopoiesis
ov.datasets.hematopoiesis_raw()——Raw hematopoiesis
ov.datasets.sc_ref_Lymph_Node()~10,000~15,000Lymph node reference
ov.datasets.bhattacherjee()~5,000~2,000Mouse PFC cocaine study
ov.datasets.human_tfs()——Human TF list (DataFrame)

RNA Velocity & Trajectories

FunctionCellsGenesDescription
ov.datasets.dentate_gyrus()18,21327,998Dentate gyrus (loom)
ov.datasets.dentate_gyrus_scvelo()2,93013,913DG subset from scVelo
ov.datasets.zebrafish()4,18116,940Zebrafish developmental
ov.datasets.pancreatic_endocrinogenesis()——Pancreatic epithelial
ov.datasets.pancreas_cellrank()2,93013,913Pancreas cellrank benchmark
ov.datasets.scnt_seq_neuron_splicing()13,47644,021scNT-seq neuron splicing
ov.datasets.scnt_seq_neuron_labeling()3,06024,078scNT-seq neuron labeling
ov.datasets.sceu_seq_rpe1()~2,930~13,913scEU-seq RPE1
ov.datasets.sceu_seq_organoid()3,8319,157scEU-seq organoid
ov.datasets.haber()7,21627,998Intestinal epithelium
ov.datasets.chromaffin()——Chromaffin cell lineage
ov.datasets.hg_forebrain_glutamatergic()1,72032,738Human forebrain
ov.datasets.toggleswitch()2002Two-gene simulation

Spatial & Multiome

FunctionDescription
ov.datasets.seqfish()SeqFISH spatial transcriptomics
ov.datasets.multi_brain_5k()10x E18 mouse brain multiome (MuData)

Bulk RNA-seq & Deconvolution

FunctionDescription
ov.datasets.burczynski06()UC/CD PBMC bulk (127 samples)
ov.datasets.moignard15()Embryo hematopoiesis qRT-PCR
ov.datasets.decov_bulk_covid_bulk()COVID-19 PBMC bulk
ov.datasets.decov_bulk_covid_single()COVID-19 PBMC single-cell ref

Synthetic

FunctionDescription
ov.datasets.create_mock_dataset()Configurable synthetic scRNA-seq
ov.datasets.blobs()Gaussian blob clusters

Mock Data Generation

Use create_mock_dataset() when you need data without network access or for pipeline testing:

import omicverse as ov

# Basic mock dataset
adata = ov.datasets.create_mock_dataset(
    n_cells=2000,
    n_genes=1500,
    n_cell_types=6,
    with_clustering=False,
    random_state=42,
)
# adata.obs: cell_type, sample_id, condition, tissue
# adata.var: gene_symbols, highly_variable

# With full preprocessing (normalized, PCA, UMAP, leiden)
adata = ov.datasets.create_mock_dataset(
    n_cells=5000,
    n_genes=3000,
    n_cell_types=10,
    with_clustering=True,
)

Features:

  • Negative binomial expression distribution
  • Cell-type-specific marker genes (2-5x expression multiplier)
  • Gene names: Gene_0001, Gene_0002, ...
  • with_clustering=True adds: normalization, HVG, scaling, PCA, UMAP, leiden

Predefined Gene Set Signatures

Pre-loaded GMT files for common scoring tasks:

from omicverse.datasets import predefined_signatures, load_signatures_from_file

# Available signature keys
print(list(predefined_signatures.keys()))
# ['cell_cycle_human', 'cell_cycle_mouse', 'gender_human', 'gender_mouse',
#  'mitochondrial_genes_human', 'mitochondrial_genes_mouse',
#  'ribosomal_genes_human', 'ribosomal_genes_mouse',
#  'apoptosis_human', 'apoptosis_mouse',
#  'human_lung', 'mouse_lung', 'mouse_brain', 'mouse_liver', 'emt_human']

# Load a signature → dict[str, list[str]]
cell_cycle = load_signatures_from_file(predefined_signatures['cell_cycle_human'])
# {'S_genes': ['MCM5', 'PCNA', ...], 'G2M_genes': ['HMGB2', 'CDK1', ...]}

# Use with scoring
import scanpy as sc
sc.tl.score_genes_cell_cycle(adata, s_genes=cell_cycle['S_genes'],
                              g2m_genes=cell_cycle['G2M_genes'])

Critical API Reference

# CORRECT: use ov.datasets for standard benchmarks
adata = ov.datasets.pbmc3k()

# WRONG: manually downloading what's already built-in
# import urllib.request
# urllib.request.urlretrieve('https://...', 'pbmc3k.h5ad')  # unnecessary!
# adata = ov.read('pbmc3k.h5ad')

# CORRECT: pbmc3k(processed=True) for pre-processed version
adata = ov.datasets.pbmc3k(processed=True)

# WRONG: loading raw then manually preprocessing for a demo
# adata = ov.datasets.pbmc3k()
# sc.pp.normalize_total(adata)  # unnecessary if you just need a quick demo
阅读完整来源 ↗