Runs GPU-accelerated RCTD cell type deconvolution on spatial transcriptomics data via Python.
npx skills add https://github.com/cpanse/skills --skill rctd-py-cpanse
Please help me install this Agent Skill. Skill: rctd-py Source: https://github.com/cpanse/skills/tree/main/single-cell-ml/skills/rctd-py Command: npx skills add https://github.com/cpanse/skills --skill rctd-py-cpanse
rctd run --mode doublet --umi-min 20, then integrates the resulting cell type weights back into a Seurat object for spatial visualization. ## Quick Start Ask the agent to run rctd-py deconvolution on your spatial.h5ad and reference.h5ad files in doublet mode on the GPU and integrate the results back into your Seurat object.High-intent search queries and answers about installing and using this skill.
Install PyTorch with CUDA support, then install rctd-py and run `rctd run spatial.h5ad reference.h5ad --mode doublet --device cuda`. Validate inputs first with `rctd validate`, which requires no GPU.
rctd-py is a Python/PyTorch reimplementation of RCTD that runs 4-41x faster than R spacexr with 99.7% concordance. It supports GPU acceleration and h5ad workflows, while spacexr runs in R on CPU.
Use doublet mode for Xenium, MERFISH, and Slide-seq data, assigning 1-2 cell types per spot. Use full mode for Visium and VisiumHD where spots contain multiple cells, and multi mode for dense data with high contamination.
Reduce the batch size with `--batch-size 5000` or lower, and switch to `--dtype float32` to halve VRAM usage. Default settings use roughly 4 GB VRAM for 45 cell types, so most GPUs handle standard workloads.
Yes, GPU nodes lack Dev/* modules, so run rctd-py via `uv run rctd` from the local repo at ~/git/rctd-py. Perform any R-based reference conversion as a separate job on the employee partition first.
In R, use anndataR's as_anndata with x_layer set to counts, optionally downsampling to about 500 cells per type, then write_h5ad. Ensure .X contains raw integer counts and .obs has a cell_type column.
Lord1EgyptAnalyze spatially-resolved transcriptomics data with gene expression and spatial coordinates.
ya-wayEstimate cell-type proportions in spatial transcriptomics spots using scRNA-seq references.
chenyhvvvvAutomates RCTD cell-type deconvolution on spatial transcriptomics data, producing per-spot proportions and dominant celltype labels.
ya-wayDetect and remove doublets in single-cell RNA-seq data using Scrublet, DoubletFinder, and scDblFinder.