Table 1 Single-cell and Spatial integration strategies

Deconvolution

Spotlight [82]

scRNA-seq and spatial barcoding

Seeded non-negative

matrix factorization regression

High accuracy

Does not incorporate capture location information when modeling spatial decomposition

Spatial DWLS [83]

scRNA-seq and spatial barcoding

Weighted least squares

Higher accuracy and faster

Greater bias in estimating rare cell types

Stereo scope [84]

scRNA-seq and spatial barcoding

Problematic modelling: negative binomial distribution

Utilizes complete expression profiles to achieve greater accuracy

Requires deep sequencing depth

Robust cell-type decomposition (RCTD) [78]

scRNA-seq and spatial barcoding

Poisson distribution with maximum likelihood estimator

Systematically models the platform effect

Assumes that platform effects are shared among cells

Cell2location [85]

scRNA-seq and spatial barcoding

Hierarchical Bayesian framework

Infers the absolute number of cell types for each capture location

The hyperparameters provided by the user are generally unknown

Multimodal

intersection analysis

(MIA) [86]

scRNA-seq and

region-specific genes from spatial barcoding

Enrichment analysis

Infers the enrichment of specific cell types in a given tissue region by computing the degree of overlap between genes specifically mapped to that region.

The ST array is not large enough to cover the entire tissue, nor does each spot achieve single-cell resolution.

Conditional autoregressive-based deconvolution(CARD [86]

scRNA-seq and spatial transcriptomic data

Conditional autoregressive

Combines spatial transcriptome data with the cell type-specific reference base matrix derived from the scRNA-seq data through non-negative matrix factorization

Relies on the CAR hypothesis of the spatial structure similarity of the cell type groups at the same time.

Tangram [87]

scRNA-seq and

spatial transcriptomic data

Non-convex optimization using deep learning methods

for spatial alignment

Compatible with capture and image-based ST data

Gene expression can be less accurately predicted from histology images if the cells cannot be segmented.

DestVI [88]

Spatial transcriptomic data

Problematic method for multi-resolution analysis

Identifies important cell type-specific changes in intracellular transcriptome variation.

Excessive dimensions may result in overfitting during the deconvolution process

DSTG [89]

scRNA-seq and spatial barcoding

Semi-supervised GCN

Greater accuracy than that of benchmarked tools

Highly dependent on the quality of the link graph that models the GCN

Sepal [90]

scRNA-seq

Diffusion model

Can detect genes with irregular spatial patterns

Has CPU parallelization, but no GPU acceleration

SPARK [91]

scRNA-seq

Generalized linear spatial models

Low false discovery rate

The hyperparameters need to be reoptimized in different datasets

Mapping

PCI Seq [92]

scRNA-seq and HPRI

Problematic model

Sequencing physically interacting cells

The interacting cells cannot be too similar in transcription.

Harmony [93]

scRNA-seq and HPRI

Maximum diversity clustering and mixture model based batch correction

Can impute low abundant genes with high accuracy

The embeddings lack biological interpretability

LIGER [94]

scRNA-seq and HPRI

Integrative NMF

The LIGER algorithm captures both common cellular traits across datasets and unique features of each dataset.

Memory intensive compared to benchmarked tools

Seurat V3.0 [95]

scRNA-seq and HPRI

Analysis pipelines with integrated algorithms

A comprehensive data analysis pipeline

Only available for certain types of ST platforms

SpaGE [96]

scRNA-seq and HPRI

Domain adaptation model that aligns ST and scRNA-seq data to a common space

Less memory usage and faster than benchmarked tools in large datasets

The model includes only genes shared by both datasets.

StPlus [97]

scRNA-seq and HPRI

Based on the reference sequence

Increased accuracy, reduced time and memory consumption

Applied only to image-based sequencing data

CellTrek [98]

scRNA-seq and spatial transcriptomic data

Through co-embedding and metric learning approaches.

Maps single cells in diverse tissue types with high accuracy

Tissue areas with low density may exhibit sparse cell mapping

GimVI [99]

Spatial transcriptomic data

scRNA-seq

Variational autoencoders

Generates platform-specific patterns to improve biological interpretation

Slower than benchmarked tools

Spatially informed ligand–receptor analysis

Giotto [100]

HPRI or spatial barcoding

A toolbox containing integrated algorithms from multiple studies

Offers comprehensive pipelines for ST data analysis

Only available for some ST platforms

SpaOTsc [101]

scRNA-seq and HPRI

Structured optimal transport model

The majority of cells can be mapped accurately using a small number of genes.

Ignores the possible time delay associated with cell-to-cell communication

ProximID [102]

Cluster label permutations

Spatial transcriptomic data

Does not require physical separation of the cells in FISH images

Interactions that are not physically attached cannot be detected.

MISTy [103]

Multi-view framework to dissect effects related to cell-cell interaction,

Spatial transcriptomic data

Does not require cell type annotation

The extracted interactions cannot be directly considered as causal.

stLearn [104]

A toolbox containing integrated algorithms from multiple studies

Spatial transcriptomic data

A streamlined approach from raw inputs to in-depth downstream analysis

Only compatible with certain ST platforms

GCNG [105]

GCN

Spatial transcriptomic data

Infers novel CCls and predicts novel functional genes.

A re-optimization of the hyperparameters is required when applying the model to different datasets.

CSOmap [106]

Reconstructs cellular spatial organization based on cell- cell affinity by ligand-receptor interactions

scRNA-seq

Inference of cell-cell interactions does not require pre-definition of tissue shape

The extracted spatial structure is a pseudo-space structure