.compute_weighted_scores

This function uses an improved weighting scheme that considers:<ol>
<li>Expression level (log-normalized)</li>
<li>Detection rate (binary: above min_expression threshold)</li>
<li>Marker specificity (how unique is this marker to this cell type)</li>
<li>Expression variability (CV-based: more variable genes are more discriminative)</li>
</ol>

internal

Annotates single-cell and spatial-transcriptomic (ST) data using context-matching marker datasets. It creates a unified marker list (`Markers_list`) from multiple sources: built-in curated databases ('Cellmarker2', 'PanglaoDB', 'scIBD', 'TCellSI', 'PCTIT', 'PCTAM'), Seurat objects with cell labels, or user-provided Excel tables. SlimR first uses adaptive machine learning for parameter optimization, and then offers two automated annotation approaches: 'cluster-based' and 'per-cell'. Cluster-based annotation assigns one label per cluster, expression-based probability calculation, and AUC validation. Per-cell annotation assigns labels to individual cells using three scoring methods with adaptive thresholds and ratio-based confidence filtering, plus optional UMAP spatial smoothing, making it ideal for heterogeneous clusters and rare cell types. The package also supports semi-automated workflows with heatmaps, feature plots, and combined visualizations for manual annotation. For more details, see Kabacoff (2020, ISBN:9787115420572).

Zhao qing Wang

SlimR

.compute_weighted_scores: Compute weighted scores for per-cell annotation

Description

Usage

Arguments