Registered Functions — GPU Support Overview¶
Legend¶
- Supported: GPU acceleration available (native, parameter-enabled, or auto-enabled by environment).
- Not supported: CPU-only implementation.
- To enable GPU, initialize the environment first if needed:
ov.settings.gpu_init()(RAPIDS) orov.settings.cpu_gpu_mixed_init()(mixed). Actual GPU usage also depends on whether dependencies are installed with GPU support.
Settings¶
-
ov.settings.gpu_init: Initialize RAPIDS/GPU mode. -
ov.settings.cpu_gpu_mixed_init: Initialize CPU–GPU mixed mode.
Preprocessing (ov.pp)¶
-
ov.pp.anndata_to_GPU: Move AnnData to GPU (RAPIDS). -
ov.pp.anndata_to_CPU: Move data back to CPU. -
ov.pp.preprocess: End-to-end preprocessing (gpu rapids).mode='shiftlog|pearson'- normalize_total/log1p (gpu rapids).
- HVGs = pearson_residuals (gpu rapids).
mode='pearson|pearson'- normalize_pearson_residuals (gpu rapids).
- HVGs = pearson_residuals (gpu rapids).
-
ov.pp.scale: Scaling (gpu rapids). -
ov.pp.pca: PCA (gpu rapids | cpu-gpu-mixed[torch|mlx]). -
ov.pp.neighbors: KNN graph (bymethod).-
method='umap'(UMAP-based neighbor estimation, CPU). -
method='gauss'(Gaussian kernel, CPU). -
method='rapids'(gpu rapids).
-
-
ov.pp.umap: UMAP (by implementation).- Scanpy UMAP (
settings.mode='cpu'). - RAPIDS UMAP (
settings.mode='gpu', gpu rapids). - PyMDE/torch path (
settings.mode='cpu-gpu-mixed', cpu-gpu-mixed[torch]).
- Scanpy UMAP (
-
ov.pp.qc: Quality control (gpu rapids | cpu-gpu-mixed[torch]). -
ov.pp.score_genes_cell_cycle: Cell cycle scoring. -
ov.pp.sude: SUDE dimensionality reduction (cpu-gpu-mixed[torch]).
Utils (ov.utils)¶
-
ov.utils.mde: Minimum Distortion Embedding (all[torch]). -
ov.utils.cluster: Multi-algorithm clustering (per algorithm below).- Leiden (cpu[igraph]cpu-gpu-mixed[pyg]).
- Louvain (Scanpy, CPU).
- KMeans (scikit-learn, CPU).
- GMM/mclust (scikit-learn, CPU).
- mclust_R (R package mclust, CPU).
- schist (schist library, CPU).
- scICE (currently invoked with
use_gpu=False).
-
ov.utils.refine_label: Neighborhood voting label refinement. -
ov.utils.weighted_knn_trainer: Train weighted KNN. -
ov.utils.weighted_knn_transfer: Weighted KNN label transfer.
Single-cell (ov.single)¶
-
ov.single.batch_correction: Batch correction (per method below).- harmony (Harmony, CPU).
- combat (Scanpy Combat, CPU).
- scanorama (Scanorama, CPU).
- scVI (all[torch]).
- CellANOVA (CPU).
-
ov.single.MetaCell: SEACells (all[torch]). -
ov.single.TrajInfer: Trajectory inference (per method below).- palantir (CPU).
- diffusion_map (CPU).
- slingshot (CPU).
-
ov.single.Fate: TimeFateKernel (all[torch]). -
ov.single.pyCEFCON: CEFCON driver discovery (all[torch]). -
ov.single.gptcelltype_local: Local LLM annotation (all[torch]). -
ov.single.cNMF: cNMF (CPU implementation). -
ov.single.CellVote: Multi-method voting. -
scsa_anno(SCSA, CPU). -
gpt_anno(online GPT, CPU/network). -
gbi_anno(GPTBioInsightor, CPU/network). -
popv_anno(PopV, CPU). -
ov.single.gptcelltype: Online GPT annotation. -
ov.single.mouse_hsc_nestorowa16: Load dataset. -
ov.single.load_human_prior_interaction_network: Load prior network. -
ov.single.convert_human_to_mouse_network: Cross-species symbol conversion.
Spatial (ov.space)¶
-
ov.space.pySTAGATE: STAGATE spatial clustering (all[torch]). -
ov.space.clusters: Multi-method spatial clustering (per method below).- STAGATE (all[torch]).
- GraphST (all[torch]).
- CAST (all[torch]).
- BINARY (all[torch]).
-
ov.space.merge_cluster: Merge clusters. -
ov.space.Cal_Spatial_Net: Build spatial neighbor graph. -
ov.space.pySTAligner: STAligner integration (all[torch]). -
ov.space.pySpaceFlow: SpaceFlow spatial embedding (all[torch]). -
ov.space.Tangram: Tangram deconvolution (per mode below).-
mode='clusters'(all[torch]). -
mode='cells'(all[torch]).
-
-
ov.space.svg: Spatially variable genes (stats-based, not explicit GPU). -
ov.space.CAST: CAST integration (all[torch]). -
ov.space.crop_space_visium: Crop spatial image/coordinates. -
ov.space.rotate_space_visium: Rotate spatial image/coordinates. -
ov.space.map_spatial_auto: Auto mapping (per method below).-
method='torch'(all[torch]). -
method='phase'(NumPy, CPU). -
method='feature'(feature-based matching, CPU). -
method='hybrid'(hybrid pipeline, CPU).
-
-
ov.space.map_spatial_manual: Manual offset mapping. -
ov.space.read_visium_10x: Read Visium data. -
ov.space.visium_10x_hd_cellpose_he: H&E segmentation (gpu=True). -
ov.space.visium_10x_hd_cellpose_expand: Label expansion. -
ov.space.visium_10x_hd_cellpose_gex: GEX segmentation/mapping (gpu=True). -
ov.space.salvage_secondary_labels: Merge labels. -
ov.space.bin2cell: Bin-to-cell conversion.
External (ov.external)¶
-
ov.external.GraphST.GraphST: GraphST (devicesupports GPU). -
ov.bulk.pyWGCNA: WGCNA (CPU implementation).
Plotting (ov.pl)¶
-
ov.pl.*(_single/_bulk/_density/_dotplot/_violin/_general/_palette): plotting APIs.
Bulk (ov.bulk)¶
-
ov.bulk.*(_Deseq2/_Enrichment/_combat/_network/_tcga): statistics, enrichment, and network analysis.