What problem does it solve? Paired multimodal single-cell datasets (10x Multiome, CITE-seq) require coordinating many tools across RNA and ATAC modalities, and it is easy to apply steps in the wrong order or pick the wrong integration method. This Skill acts as the entry-point orchestrator that chains preprocessing, batch integration, differential accessibility, peak-gene linkage, and visualisation across the scverse Python ecosystem. ## Core Features & Use Cases - Decision-tree routing: Maps your starting point (Cell Ranger ARC output, h5mu, or fragments-only) to the correct leaf skill for loading and preprocessing each modality. - Canonical workflow: Provides an end-to-end code path covering MuData loading, scanpy RNA preprocessing, SnapATAC2 ATAC preprocessing, Harmony batch correction, WNN neighbors, Leiden clustering, and differential accessibility. - Integration guidance: Explains when to stay on WNN + Harmony versus switching to scvi-tools MultiVI for joint deep-learning embeddings, plus CITE-seq and cross-modality pitfalls. - Use Case: Given a 10x Multiome experiment with multiple samples, follow the orchestrator to produce a batch-corrected joint UMAP, per-cluster marker peaks, and peak-to-gene cis-regulatory links saved as an h5mu file. ## Quick Start Ask the assistant to run the muon multimodal workflow on your 10x Multiome filtered_feature_bc_matrix.h5 and fragments file to produce an integrated h5mu with marker peaks.
