Introduction to Multiplexed In Situ Technologies
Researchers Hartman and Satija from the New York Genome Center and New York University have released a significant preprint entitled "Comparative Analysis of Multiplexed In Situ Gene Expression Profiling Technologies.” The study presents a comparative analysis of six in situ gene expression profiling methods.
They assessed these technologies for their sensitivity, particularly in detecting unique molecules per cell, and for their specificity, focusing on the impact of off-target molecular artifacts. In addition, the researchers developed new metrics to control for these artifacts. Their findings highlight the variability among different in situ methods and emphasize the importance of considering molecular false positives in spatially-aware differential expression analysis.
Technological Insights and Comparative Analysis
This study focused on benchmarking various in situ spatial technologies used for multiplexed gene expression profiling, using datasets from full mouse brain sections. The technologies compared include Xenium (10x Genomics), MERSCOPE (Vizgen), Molecular Cartography (Resolve Biosciences), MERFISH, STARmap PLUS, and EEL FISH. The researchers aimed for a fair comparison by utilizing outputs provided by the dataset creators, which included gene-cell expression matrices and coordinates for each segmented cell.
The initial analysis measured reproducibility and average counts and features identified in each cell. High reproducibility was observed across technologies. However, sensitivity varied, with no correlation found between the average number of transcripts detected per cell and total panel size. The study also assessed the ability of these technologies to categorize cell types in mouse brain sections, finding success in identifying broad cell groups but also noting non-specific expression of marker genes. Higher-resolution subsets were not benchmarked due to differences in panel size and composition across technologies.
A comparison of molecular sensitivity between in situ technologies and scRNA-seq revealed that in situ methods generally showed higher molecular counts. However, this could be partly due to non-specific signals inflating gene expression counts. The study also observed a higher increase in molecular counts for lowly expressed genes from in situ techniques, suggesting the presence of a uniform background effect.
Non-specificity was further examined through a 'mutually exclusive co-expression rate' (MECR), which varied across technologies. The study also explored potential sources of non-specific signals including off-target probe binding and the accuracy of molecular assignment to cells. Variations in cell segmentation quality across technologies affected molecular assignment accuracy. A uniform segmentation processing pipeline was applied to reprocess datasets, revealing significant differences in sensitivity and specificity.
Further, the researchers examined how spatial location influences gene expression within a cell type. They found discrepancies in differentially expressed genes when comparing in situ datasets with scRNA-seq data, suggesting non-specific molecular signals and potential misassignment of molecules from in situ datasets. This was particularly evident in genes that appeared upregulated in certain cell types due to molecular misassignment.
Main Findings
Researchers Hartman and Satija from the New York Genome Center and New York University have released a significant preprint entitled "Comparative Analysis of Multiplexed In Situ Gene Expression Profiling Technologies.” The study presents a comparative analysis of six in situ gene expression profiling methods.
They assessed these technologies for their sensitivity, particularly in detecting unique molecules per cell, and for their specificity, focusing on the impact of off-target molecular artifacts. In addition, the researchers developed new metrics to control for these artifacts. Their findings highlight the variability among different in situ methods and emphasize the importance of considering molecular false positives in spatially-aware differential expression analysis.
Technological Insights and Comparative Analysis
This study focused on benchmarking various in situ spatial technologies used for multiplexed gene expression profiling, using datasets from full mouse brain sections. The technologies compared include Xenium (10x Genomics), MERSCOPE (Vizgen), Molecular Cartography (Resolve Biosciences), MERFISH, STARmap PLUS, and EEL FISH. The researchers aimed for a fair comparison by utilizing outputs provided by the dataset creators, which included gene-cell expression matrices and coordinates for each segmented cell.
The initial analysis measured reproducibility and average counts and features identified in each cell. High reproducibility was observed across technologies. However, sensitivity varied, with no correlation found between the average number of transcripts detected per cell and total panel size. The study also assessed the ability of these technologies to categorize cell types in mouse brain sections, finding success in identifying broad cell groups but also noting non-specific expression of marker genes. Higher-resolution subsets were not benchmarked due to differences in panel size and composition across technologies.
A comparison of molecular sensitivity between in situ technologies and scRNA-seq revealed that in situ methods generally showed higher molecular counts. However, this could be partly due to non-specific signals inflating gene expression counts. The study also observed a higher increase in molecular counts for lowly expressed genes from in situ techniques, suggesting the presence of a uniform background effect.
Non-specificity was further examined through a 'mutually exclusive co-expression rate' (MECR), which varied across technologies. The study also explored potential sources of non-specific signals including off-target probe binding and the accuracy of molecular assignment to cells. Variations in cell segmentation quality across technologies affected molecular assignment accuracy. A uniform segmentation processing pipeline was applied to reprocess datasets, revealing significant differences in sensitivity and specificity.
Further, the researchers examined how spatial location influences gene expression within a cell type. They found discrepancies in differentially expressed genes when comparing in situ datasets with scRNA-seq data, suggesting non-specific molecular signals and potential misassignment of molecules from in situ datasets. This was particularly evident in genes that appeared upregulated in certain cell types due to molecular misassignment.
Main Findings
- Vizgen's MERSCOPE datasets showed the best performance, with an optimal balance between sensitivity and specificity, and a large panel size of 483 genes.
- Xenium technology, with a 247-gene panel, also performed well but had lenient segmentations that increased molecular detection at the cost of accuracy.
- Resolve Molecular Cartography and MERFISH datasets showed similar sensitivity but differed in panel sizes (99 genes for Molecular Cartography and 1147 for MERFISH).
- EEL FISH was the least sensitive technology, trading sensitivity for faster imaging time and reduced autofluorescence.
- The study's limitations include its focus on mouse brain tissue and fresh-frozen samples, limiting its applicability to a single biological context.
- Non-specific signals in imaging-based spatial transcriptomics is largely due to misassignment of molecules, necessitating improved segmentation and computational methods to address this issue.