Guide Claude through omicverse's single-cell clustering workflow, covering preprocessing, QC, multimethod clustering, topic modeling, cNMF, and cross-batch integration as demonstrated in t_cluster.ipynb and t_single_batch.ipynb.
This skill distills the single-cell tutorials t_cluster.ipynb and t_single_batch.ipynb. Use it when a user wants to preprocess an AnnData object, explore clustering alternatives (Leiden, Louvain, scICE, GMM, topic/cNMF models), and evaluate or harmonise batches with omicverse utilities.
omicverse as ov, scanpy as sc, and plotting helpers (scvelo as scv when using dentate gyrus demo data).ov.plot_set() or ov.utils.ov_plot_set() so figures adopt omicverse styling before embedding plots.scv.datasets.dentategyrus(); for integration, read provided .h5ad files via ov.read() and set adata.obs['batch'] identifiers for each cohort.adata.X = adata.X.astype(np.int64) when required for QC steps.ov.pp.qc(adata, tresh={'mito_perc': 0.2, 'nUMIs': 500, 'detected_genes': 250}, batch_key='batch') to drop low-quality cells and inspect summary statistics per batch.adata.write_h5ad(...)) so users can resume from clean checkpoints.ov.pp.preprocess(adata, mode='shiftlog|pearson', n_HVGs=3000, batch_key=None) to normalise, log-transform, and flag highly variable genes; assign adata.raw = adata and subset to adata.var.highly_variable_features for downstream modelling.ov.pp.scale(adata)) and compute PCA scores with ov.pp.pca(adata, layer='scaled', n_pcs=50). Encourage reviewing variance explained via ov.utils.plot_pca_variance_ratio(adata).sc.pp.neighbors(adata, n_neighbors=15, n_pcs=50, use_rep='scaled|original|X_pca') or ov.pp.neighbors(...).ov.utils.cluster(adata, method='leiden'|'louvain', resolution=1), ov.single.leiden(adata, resolution=1.0), or ov.pp.leiden(adata, resolution=1); remind users that resolution tunes granularity.# Before clustering: check neighbors graph exists
if 'neighbors' not in adata.uns:
if 'X_pca' in adata.obsm:
ov.pp.neighbors(adata, n_neighbors=15, use_rep='X_pca')
else:
raise ValueError("PCA must be computed before neighbors graph")
# Before plotting by cluster: check clustering was performed
if 'leiden' not in adata.obs:
ov.single.leiden(adata, resolution=1.0)
ov.pl.embedding(adata, basis='X_umap', color=['clusters','leiden'], frameon='small', wspace=0.5) and confirm cluster separation. Always check that columns in color= parameter exist in adata.obs before plotting.model = ov.utils.cluster(adata, method='scICE', use_rep='scaled|original|X_pca', resolution_range=(4,20), n_boot=50, n_steps=11) and inspect stability via model.plot_ic(figsize=(6,4)) before selecting model.best_k groups.ov.utils.cluster(..., method='GMM', n_components=21, covariance_type='full', tol=1e-9, max_iter=1000) for model-based assignments.LDA_obj = ov.utils.LDA_topic(...), review LDA_obj.plot_topic_contributions(6), derive cluster calls with LDA_obj.predicted(k) and optionally refine using LDA_obj.get_results_rfc(...).cnmf_obj = ov.single.cNMF(... components=np.arange(5,11), n_iter=20, num_highvar_genes=2000, output_dir=...), factorise (factorize, combine), select K via k_selection_plot, and propagate usage scores back with cnmf_obj.get_results(...) and cnmf_obj.get_results_rfc(...).adjusted_rand_score(adata.obs['clusters'], adata.obs['leiden']) and report metrics for each method (Leiden, Louvain, GMM, LDA variants, cNMF models) to justify chosen parameters.ov.utils.mde(...) to create MDE projections from different latent spaces (adata.obsm["scaled|original|X_pca"], harmonised embeddings, topic compositions) and plot via ov.utils.embedding(..., color=['batch','cell_type']) or ov.pl.embedding for consistent review of cluster/batch mixing.ov.single.batch_correction(adata, batch_key='batch', methods='harmony'|'combat'|'scanorama'|'scVI'|'CellANOVA', n_pcs=50, ...) sequentially to generate harmonised embeddings stored in adata.obsm (X_harmony, X_combat, X_scanorama, X_scVI, X_cellanova). For scVI, mention latent size (n_latent=30) and gene_likelihood="nb"; for CellANOVA define control pools via control_dict.ov.utils.mde and visualise batch vs cell_type to check mixing and conservation.adata.write_h5ad('neurips2021_batch_all.h5ad', compression='gzip')) and reload when necessary.scib_metrics.benchmark.Benchmarker with embeddings list (["X_pca", "X_combat", "X_harmony", "X_cellanova", "X_scanorama", "X_mira_topic", "X_mira_feature", "X_scVI"]) to compute batch-vs-biology trade-offs via bm.benchmark() and summarise with bm.plot_results_table(min_max_scale=False).adata.raw captures the unscaled log-normalised matrix before subsetting to HVGs.use_rep='scaled|original|X_pca' strings exist in adata.obsm prior to clustering; rerun preprocessing if missing.n_iter, components, or latent dimensions for smaller datasets.'neighbors' in adata.unsadata.obsclusters."batch and cell_type, and benchmark the embeddings."t_cluster.ipynbt_single_batch.ipynbreference.md