Seurat

Implements the SCTransform method, which uses regularized negative binomial regression to normalize raw Unique Molecular Identifier (UMI) counts. This approach effectively removes the confounding influence of sequencing depth while preserving biological heterogeneity. The framework also supports log-normalization for compatibility with legacy workflows, ensuring variance stabilization across the count matrix.

Performs Principal Component Analysis (PCA) on highly variable genes (HVGs) to capture the dominant axes of biological variation. Seurat constructs a Shared Nearest Neighbor (SNN) graph in PCA space, which serves as the input for non-linear visualization techniques like t-SNE and UMAP. This graph-based approach ensures that the local neighborhood structure of cells is preserved in low-dimensional embeddings.

Leverages community detection algorithms, primarily the Louvain and Leiden algorithms, to partition cells into transcriptionally distinct clusters. By operating on the SNN graph, the method identifies groupings at multiple resolutions without requiring a priori knowledge of the number of populations. This modularity optimization approach is robust to noise and scalable to millions of cells.

Uses a canonical correlation analysis (CCA)-based workflow to identify anchors—mutual nearest neighbors across datasets—that represent cells in a shared biological state. These anchors are used to compute a transformation matrix that harmonizes distinct experiments, effectively correcting for batch effects while maintaining condition-specific biological variation. This enables joint analysis of cross-technology or cross-laboratory data.

Provides statistical frameworks for identifying cluster-specific marker genes and performing differential expression testing between conditions. The FindMarkers function supports multiple test methodologies, including the Wilcoxon Rank Sum test and logistic regression, to pinpoint genes that define cellular identity or disease state. Results are structured for immediate downstream pathway enrichment analysis.

Extends the core framework to analyze multimodal single-cell data, including CITE-seq (RNA + surface proteins) and scATAC-seq (chromatin accessibility), using a weighted nearest neighbor (WNN) approach. For spatial transcriptomics, Seurat integrates gene expression with tissue coordinate information, enabling spatially variable gene detection and the mapping of reference cell types onto histological sections.

A sparse numerical matrix where rows represent genes and columns represent cell barcodes, storing the number of transcripts detected per gene in each cell. This is the primary input for Seurat's CreateSeuratObject() function.

• Typically stored in formats like .mtx or .h5
• Extremely sparse, often with >90% zero entries due to biological and technical dropout
• Seurat stores this in the counts assay slot

The initial filtering step that removes low-quality cells based on metrics like total UMI counts, number of genes detected, and mitochondrial read fraction. Seurat facilitates this through metadata filtering.

• High mitochondrial percentage (>20%) indicates dying or damaged cells
• Low gene counts suggest empty droplets or ambient RNA contamination
• Doublet removal is often performed computationally after initial QC

The process of scaling raw count data to adjust for differences in sequencing depth and capture efficiency between cells. Seurat's default NormalizeData() uses a global-scaling log-normalization method.

• Transforms counts to log(CPM/100 + 1) by default
• Enables valid cross-cell comparisons of gene expression
• Alternative methods like SCTransform use regularized negative binomial regression for more effective variance stabilization

Genes exhibiting greater expression variance across cells than expected by technical noise. Seurat's FindVariableFeatures() selects these as informative features for downstream dimensionality reduction.

• Default method identifies the top 2,000 HVGs
• Focuses analysis on genes with the strongest biological signal
• Reduces computational burden by ignoring stochastically expressed or uniformly low genes

Mathematical transformation of high-dimensional single-cell data into a lower-dimensional space. Seurat runs Principal Component Analysis (PCA) on scaled HVGs, then uses the significant PCs to compute a UMAP or t-SNE for visualization.

• PCA identifies the axes of maximum variance
• UMAP preserves both local and global data structure for clustering
• The ElbowPlot() function helps determine the number of significant PCs to retain

An unsupervised method that partitions cells into groups by constructing a Shared Nearest Neighbor (SNN) graph and applying community detection. Seurat's FindClusters() uses the Louvain or Leiden algorithm.

• A resolution parameter controls the granularity of clustering
• Clusters represent transcriptionally distinct cell populations
• Results are stored in the seurat_clusters metadata column