Evaluation of spatial transcriptomics (ST) tools across various metrics and conditions.
<p><b>(A)</b> Bar plots showing Pearson correlation coefficients (PCCs) and Root Mean Square Errors (RMSEs) between predicted and true proportions in synthetic ST datasets under different spot resolution levels (<i>Lambda</i> = 5, 10, 15, and 20). The <i>Lambda<...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | , , , , , , , , |
| Published: |
2025
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p><b>(A)</b> Bar plots showing Pearson correlation coefficients (PCCs) and Root Mean Square Errors (RMSEs) between predicted and true proportions in synthetic ST datasets under different spot resolution levels (<i>Lambda</i> = 5, 10, 15, and 20). The <i>Lambda</i> value represents the expected cell counts in spots, modeled by a Poisson distribution. Error bars indicate variability, determined through a permutation strategy repeated 100 times, where 2,000 spots were down-sampled, and PCC and RMSE were calculated for each iteration. <b>(B)</b> Violin combined with box plots showing the PCCs and RMSEs between estimated and actual cellular proportions. Black lines within the boxes indicate medians. Thirty-two simulated ST datasets, together with their corresponding single-cell (SC) datasets, were retrieved from the study by Li and colleagues [<a href="http://www.plosbiology.org/article/info:doi/10.1371/journal.pbio.3003477#pbio.3003477.ref028" target="_blank">28</a>]. <b>(C)</b> Kullback–Leibler (<i>KL</i>) divergence metrics for spatial cell charting methods, comparing each cell type against synthetic ST dataset references. <b>(D)</b> Grouped bar plots showing cosine similarities between recovered spot expressions post-SC mapping and the original synthetic ST dataset expressions. Error bars represent values obtained via permutation strategy, with 2,000 spots down-sampled and cosine similarities calculated across 100 iterations. <b>(E)</b> Grouped bar plots showing the conservation of highly variable genes (HVGs) between recovered spot expressions and synthetic ST dataset expressions following SC mapping. <b>(F)</b> Bar plot showing the average Jaccard index for cell type consistency in spots, comparing predictions from our score-guided mapping accuracy (SGMA; see “Materials and methods”) strategy against true cell types in synthetic ST datasets. <b>(G)</b> Box plot displaying SGMA indexes for cell type consistency in spots, comparing predictions using mapping tools (Cell2Spatial, CytoSPACE, CellTrek, and Tangram) with true cell types in synthetic ST datasets, based on the SGMA strategy (see “Materials and methods”). <b>(H)</b> Scatter plot mapping a subset of mouse brain SC data to ST spots under unmatched conditions (where cell types from SC data represent only a subset of those in ST datasets). Each dot corresponds to an individual cell, with cell types color-coded. <b>(I)</b> Violin combined with box plots showing the PCC and RMSE between estimated and actual cell counts in spots. Simulated ST datasets, together with their corresponding SC datasets, were retrieved from the study by Li and colleagues [<a href="http://www.plosbiology.org/article/info:doi/10.1371/journal.pbio.3003477#pbio.3003477.ref028" target="_blank">28</a>]. <b>(J)</b> Spatial feature plots illustrating inferred cell counts in spots using the DAPI channel in fluorescence images (counted via Squidpy [<a href="http://www.plosbiology.org/article/info:doi/10.1371/journal.pbio.3003477#pbio.3003477.ref029" target="_blank">29</a>]), Cell2Spatial, and CytoSPACE. The top panel represents a coronal section of the mouse brain, while the bottom panel represents the whole brain. Color shading indicates different cell counts in spots. <b>(K, L)</b> Density plots combined with fitted lines, showing the consistency between cell counts estimated by Cell2Spatial or CytoSPACE and those counted using DAPI via the Squidpy tool [<a href="http://www.plosbiology.org/article/info:doi/10.1371/journal.pbio.3003477#pbio.3003477.ref029" target="_blank">29</a>]. “<i>R”</i> represents the Spearman correlation, with <i>p</i>-values obtained from the two-sided <i>t</i>-tests. <b>(M)</b> Scatter plot with line showing the influence of marker size on Cell2Spatial performance for different cell types. The marker size range is 10 to 200, with a step size of 10. <b>(N, O)</b> Scatter plots with line showing peak memory usage and elapsed time for different mapping tools under varying spot sizes, ranging from 1,000 to 10,000, with a step size of 1,000. The data underlying this figure can be found at <a href="https://doi.org/10.5281/zenodo.17212677" target="_blank">https://doi.org/10.5281/zenodo.17212677</a>.</p> |
|---|