Spatial Batch Correction

Identifies cross-sample cell pairs that are mutual nearest neighbors in a shared feature space, using these as anchors to compute correction vectors. This approach assumes that technical variation shifts the population in a consistent direction, while biological variation is preserved by only correcting within matched cell states. The method is robust to differences in cell-type composition between samples and scales efficiently to large spatial datasets.

Projects datasets into a shared low-dimensional subspace that maximizes the correlation between corresponding features across batches. In spatial contexts, CCA identifies sources of shared biological variation while isolating batch-specific noise. The corrected embeddings can then be used for downstream spatial domain detection and trajectory inference without the confounding influence of technical artifacts.

An iterative clustering and correction algorithm that uses a soft k-means clustering approach to group cells into diverse clusters, then applies cluster-specific linear correction factors. Harmony is particularly effective for spatial data because it does not assume all cell types are shared across batches, making it resilient to tissue-specific cell populations that may be absent in some samples.

Assembles a panoramic view of the cellular landscape by finding correspondences between datasets using a mutual nearest neighbors approach, then merging them with a technique analogous to image stitching. Scanorama corrects batch effects by matching overlapping cell types and applying a nonlinear transformation that smoothly blends the datasets while preserving unique biological structures present in only one sample.

Employs generalized linear models that explicitly account for the count-based, overdispersed nature of spatial transcriptomics data. Batch identity is included as a covariate in the regression, allowing the model to estimate and remove its effect on gene expression. This parametric approach provides interpretable coefficients and naturally handles the zero-inflated distributions common in spatial data.

Extends standard batch correction by incorporating spatial neighborhood information as a regularization term. The algorithm penalizes corrections that disrupt known spatial autocorrelation patterns, ensuring that the corrected expression values maintain realistic spatial gradients. This prevents over-correction that could flatten genuine biological spatial heterogeneity, such as developmental gradients or tumor-immune boundaries.

The process of combining multiple spatial transcriptomics datasets from different technologies, samples, or modalities into a unified analytical framework. Unlike non-spatial integration, these methods must preserve tissue architecture and spatial coordinates.

• Horizontal integration: Merging multiple slices from the same technology (e.g., multiple 10x Visium samples)
• Vertical integration: Aligning different data modalities (e.g., spatial transcriptomics + spatial proteomics)
• Anchor-based methods: Use shared features or cell types to learn a common latent space

The computational alignment of multiple tissue images or spatial datasets into a common coordinate system. This is a critical preprocessing step before batch correction, ensuring that the same anatomical region is compared across samples.

• Rigid registration: Translation and rotation only
• Affine registration: Adds scaling and shearing
• Non-linear/diffeomorphic registration: Warps tissue to match complex anatomical boundaries
• Tools like STUtility and PASTE perform probabilistic alignment of spatial transcriptomics spots

A statistical measure of the degree to which a variable's values at nearby locations are more similar than expected by random chance. Effective batch correction must preserve this signal while removing technical artifacts.

• Positive autocorrelation: Neighboring spots have similar expression (e.g., a tumor region)
• Negative autocorrelation: Neighboring spots are dissimilar (e.g., sharp boundaries)
• Moran's I quantifies this globally; Local Indicators of Spatial Association (LISA) identify specific clusters

The unsupervised identification of tissue regions with coherent gene expression profiles and histology. After batch correction, domain detection validates that biological structures—not batch artifacts—drive clustering.

• BayesSpace: Uses a spatial hidden Markov model with a Potts prior to encourage spatial smoothness
• SpaGCN: A graph convolutional network that integrates histology images with gene expression
• stLearn: Incorporates spatial distance and morphological similarity into clustering

A non-parametric statistical test that randomly shuffles spatial labels to generate a null distribution. This is the gold standard for validating that observed spatial patterns are not artifacts of batch effects or random chance.

• Procedure: Randomly permute spot coordinates, recalculate the statistic of interest, repeat 1,000+ times
• Application: Assess significance of spatially variable genes, cell-type colocalization, or ligand-receptor interactions
• Squidpy and Giotto provide built-in spatial permutation frameworks