Hello, I'm still the same👉 Share, summarize, and compare idle tools from time to timeToday, I'd like to "feed" you this article from the Nature journalNature Methods", researchers compared spatial transcriptome and single-cell transcriptome integration analysis tools to measure theirperformances。
Performance testing of the idle&scRNA-seq integrated analysis tool
Spatial transcriptomics methods allow us to detect RNA transcripts in space, and these methods have been used to study the spatial distribution of gene expression in a variety of tissues and organs, including the brain, heart, pancreas, and skin.On the one hand, spatial transcriptomics methods based on in situ hybridization and fluorescence microscopy (image-based) (including seqFISH, osmFISH, and MERFISH) are able to detect the spatial distribution of transcripts with high resolution and accuracy, but they can only detect the total number of RNA transcripts. On the other hand, spatial transcriptomics methods based on next-generation sequencing (sequencing-based), such as ST, 10X Visium, and Slide-seq, can capture expressed RNAs at the level of the entire transcriptome from points in space, but each point (with a radius of 10-100 µm) may contain multiple cells, which limits the spatial resolution of these methods. The limitations of these spatial transcriptomics methods hinder their ability to capture whole transcriptome-scale data at spatial single-cell resolution.
To break through the limitations of spatial transcriptomics approaches, bioinformaticians have proposed and developed a variety of integrative methods to combine spatial transcriptomics and single-cell transcriptome (scRNA-seq) dataIncludesgimVI、SpaGE、Tangram、Seurat、novoSpaRc、SpaOTscetc. There has not been an independent study that comprehensively compares these integration methods in predicting the spatial distribution of transcripts or the cell type of spots in tissue sections deconvolutionaspects of performance. As a result, theThe researchers in this study systematically benchmarked the performance of 16 integration methods using a variety of metrics.
Benchmarking process and brief characterization of the dataset used
The researchers assessed the performance of the correlation tools on multiple datasets using Pearson's correlation coefficient (PCC), structural similarity (SSIM), root-mean-square error (RMSE), and JS dispersion, while aggregating the above four metrics to define accuracy scores (AS) to simplify the assessment of the accuracy of each integration tool (higher AS values indicate better performance).
Summary of performance evaluation results
Tangram, gimVI and SpaGE outperform other integration methods in predicting spatial distribution of transcripts, where Tangram, gimVI, and SpaGE outperformed the other integrated methods in processing data generated by the 10X Visium, seqFISH, and MERFISH platforms, and Tangram and gimVI were the preferred methods for processing Slide-seq datasets.
Also, Tangram, gimVI and SpaGE outperform other integration methods in predicting the spatial distribution of transcripts in highly sparse datasets.
Cell2location, SpatialDWLS, RCTD, and STRIDE outperformed the other integration methods in predicting the cell type composition of spots.
existcomputing resourceAspects.Seuratis the most computationally efficient method for predicting the spatial distribution of transcripts;Tangram and SeuratIt is the most efficient method for dealing with cell type deconvolution.
In summary, Tangram, gimVI, and SpaGE outperformed the other integration methods in predicting the spatial distribution of transcripts, while Cell2location, SpatialDWLS, and RCTD outperformed the other integration methods in de-convolution of cell types to spots in histological sections.
probability-basedmouldMethods that incorporate negative binomial or Poisson distribution construction, such as thegimVI, Cell2location, and RCTD, which typically perform better in predicting the spatial distribution of transcripts or the cell type of de-convoluted spots。deep learningarithmeticIt is also applied to a variety of integration methods in whichTangram is one of the best performing methods for predicting the spatial distribution of undetected transcripts。
Targeting spatial transcriptomicsThe sparsity problem for expression matrices, there are multiple strategies:Researchers can increase the sequencing depth, filter spots and genes with strict cut-off values to reduce the sparsity of the filtered expression matrix, or consider applying interpolation algorithms (e.g., SAVER, MAGIC, and WEDGE) to interpolate the zero elements of the expression matrix。
Another potential application of spatial transcriptomics is the prediction of spatial proximity between two cell types in close proximity to each other'sligand-receptor interactionThe analysis tools have now been developed in a number of ways, such asSpaOTsc, Giotto, CellChat, NicheNet, ICELLNET andSingleCellSignalRetc.
//
We recommend that those interested in the details of the test refer to the original paper ~ or you can visit the following link for details
👉 /QuKunLab/SpatialBenchmarking
First Publication: National Gene Bank Life Big Data Platform
bibliography
Li, B., Zhang, W., Guo, C. et al. Benchmarking spatial and single-cell transcriptomics integration methods for transcript distribution prediction and cell type Methods (2022).
Images are from references, please contact to remove any infringement.