data-io-loading_skill

Overview

This skill provides OmicVerse-focused data I/O helpers for loading bulk, single-cell, and spatial RNA-seq formats. It replaces common Scanpy and pandas reads with ov.read() and ov.io.* readers that handle h5ad, 10x H5/MTX, Visium, Visium HD, Nanostring, and CSV/TSV robustly. Use these readers in OmicVerse projects to avoid format edge cases and gain spatial geometry and Rust-backend options.

How this skill works

The skill exposes a universal ov.read(path, backend='python') that auto-detects files and routes to specialized readers under ov.io and ov.io.spatial. Readers include read_h5ad, read_10x_h5, read_10x_mtx, spatial.read_visium, read_visium_hd (bin/seg variants), read_nanostring, and read_csv. Many readers auto-detect variants (10x v2/v3, Visium parquet/csv, Nanostring column variants) and provide options for images, scalefactors, segmentation polygons, and Rust-backed h5ad loading.

When to use it

• Loading any .h5ad, .csv, .tsv, or gzipped table into an OmicVerse workflow
• Importing 10x Genomics outputs (H5 or Matrix Market) with correct format detection
• Loading Visium spatial experiments with images, tissue positions, and scalefactors
• Working with Visium HD (binned outputs or cell-segmentation polygons)
• Reading Nanostring/CosMx SMI outputs that require column-name auto-detection

Best practices

• Always call ov.read(path) instead of sc.read or pandas.read_csv in OmicVerse projects
• Validate input paths before calling readers (file/directory existence, expected files in Space Ranger outputs)
• Use ov.io.read_h5ad(..., backed='r') or ov.read(..., backend='rust') for very large h5ad files and remember to close backed objects
• Prefer gex_only=False when you need antibody or guide features from 10x H5 files
• Install optional dependencies (geopandas, shapely) to enable polygon geometry support for Visium HD segmentation

Example use cases

• Load a small h5ad for analysis: adata = ov.read('dataset.h5ad')
• Import 10x HDF5 counts with antibody features: adata = ov.io.read_10x_h5('filtered_feature_bc_matrix.h5', gex_only=False)
• Read Space Ranger Visium with images: adata = ov.io.spatial.read_visium('outs/', load_images=True)
• Open Visium HD segmentation to get WKT polygons: adata = ov.io.read_visium_hd_seg('outs/segmented_outputs/')
• Read Nanostring CosMx outputs with counts and metadata: ov.io.read_nanostring(path, counts_file, meta_file, fov_file=None)