University of Iowa: Institute for Vision Research

The Spectacle interface is organized into three columns. (A) The left most column comprises the sidebar which contains five menu items: “Dashboard”, “About”, “Citation and Dataset Info”, “How to guide”, and “cellcuratoR”. When the “Dashboard” menu item is selected, the interactive single-cell app can be accessed. The dataset of interest can be selected within the dataset dropdown (*). The three horizontal bars (arrow) can be selected to hide/show the sidebar for more screen real estate. Within the Dashboard menu item, each reactive content space has multiple tabs that can be selected to display different visualizations. Below each tab, there is a space for reactive helper functions that provide user inputs to customize the overlying visualizations.The “About” menu item contains information about the Iowa Institute for Vision Research, which developed Spectacle and cellcuratoR. The “Citation and Dataset Info” menu item contains information about how to cite Spectacle and the datasets contained in this platform. The “How to guide” menu item contains screen shots and tutorial videos for using Spectacle. The “cellcuratoR” menu item provides information about the R package (cellcuratoR) that powers Spectacle, along with a link to the source code on GitHub.

In the dashboard menu item (arrow), once a dataset is selected, an interactive plot is displayed in the Dimensionality Reduction Tab. Cells in this plot can be colored according to their cluster classification (A), or by library composition of the cells (B) by selecting the color scheme from the drop-down helper, Dimensionality reduction plot color.

A. Expression patterns for a gene of interest can be overlaid on the dimensionality reduction plot when the “Heatmap” tab (arrow) is selected. Once the heatmap tab is selected, the “gene for heatmap” helper input appears below the plot, where the user can input (as text) the gene name to query. By default, only 100 genes are avaialble in this reactive text input box at a time. Simply start typing the name of the gene, and a string-matching search will return gene names with matches to the text input. To the right of the heatmap, a custom legend mapping color to expression value appears. The scale of expression in this legend is “TP.10K,” or transcripts per 10,000, as all datasets used in Spectacle were normalized with a scale.factor of 10000 using the NormalizeData() function.

B. Alternatively, expression can be visualized on a per-cluster level with violin plots. On the right side of the violin plots, a dendrogram depicts the relationships between each cluster. By default, the export_seurat_object function creates a dendrogram by identifying the 100 most differentially expressed genes in each cluster, finding the average expression of these genes across all clusters, and performing hierarchical clustering on the resulting expression matrix with a Canberra distance metric and complete linkage. On the right side of the plot, violins are drawn depicting expression in each cluster. If less than 25% of the cells in a cluster do not express the gene of interest above background, a vertical line is drawn representing the range of detected expression levels, but the violin distribution is not displayed. Multiple genes can be simultaneously visualized within the violin plot by adding gene names to the helper input “Violin genes.”

Clusters of interest can be re-clustered in a lower dimensional feature space within the “Recluster” tab (arrow). Once selected, the user will be prompted to choose one or more clusters of interest to recluster in the input helper selection (A). In this example, cone clusters 3 and 4 were selected with the checkbox input. The user can then initiate the reclustering by pushing the “start reclustering (slow)” action button. A progress bar will be displayed on the lower right corner of the window indicating each step of the reclustering. When complete, a plot of the reclustered cells appears. By default, these cells are clustered according to the original cluster color, as depicted in the dimensionality reduction tab on the left side of the screen. However, cells can also be colored by libraryID or by expression of a gene of interest by changing the input to “How should the cells be colored?” at the bottom of the helper box.

Reclustering parameters can be tuned after clicking the advanced settings checkbox. The user can adjust the default dimensionality reduction strategy for visualization, which includes principal component analysis (pca), t-distributed stochastic neighbor embedding (tSNE), or uniform manifold approximation and projection (UMAP). Likewise, the user can adjust parameters important in normalization, such as the number of variable features and how these features are selected. Finally, the user can adjust the granularity of the reclustered object, which can be visualized by coloring cells 'by new cluster' under the 'How should the cells be colored?' dropdown menu. Conversely, if the user selects to color reclustered cells by gene (bottom), an additional selection appears where the gene of interest can be input. Of note, the dimesnionality reduction method and granularity of the reclustered object can be modified without re-executing the reclustering process. However, adjusting the number of variable features or how they are selected requires re-clicking 'start reclustering (slow)' for the changes to take effect.

Differential expression can be conducted between clusters of cells. First, the user should select the two differential expression tabs at the top of the dashboard window (arrows). This will result in display of the differential expression helper input appearing (A). The user may adjust the logFC.threshold, which removes genes with mean expression differences between the two groups that are less than the specified threshold. Additionally the user may adjust the minimum percent threshold, which removes genes that are not expressed in at least the selected proportion of cells in either of the comparison groups. Thresholds with larger values result in faster differential expression testing with less sensitivity, while thresholds with smaller values result in slower differential expression testing with more sensitivity. The lower limit of both thresholds is set to 0.1 to minimize load on the hosted server. Next, the user selects the dataset for differential expression. If a reclustering procedure has been performed (see previous tabs), then the reclustered object is avaialble for differential expression. However in most cases, the default object containing all cells is most appropriate for differential expression testing. Next, the user should select what groups will be compared with the differential expression (horizontal arrows). If the user selects the “predefined clusters” radio button, an additional differential expression helper will be displayed on the right side of the screen where the user can select one or more clusters of cells to be compared. In this example, cone cluster 3 is being compared to cone cluster 4. When complete, the user then may click the “Start Differential Expression Analysis” action button, prompting differential expression to start. A progress bar will appear in the lower right corner of the window that displays the progress of the differential expression testing. When complete, the differential expression results are displayed both visually and in table format. In (C), the average log fold-change (y-axis) and delta.percent (x-axis) of the differentially expressed genes are displayed, where each point depicts the differential expression results of a single gene. The delta.percent variable depicts the percentage of cells in group 1 that express a gene above background minus the percent of cells in group 2 that express the gene above background. For example, transferrin (TF) is expressed in 80.9% of cones in Cluster 4 and 6.4% of cones in Cluster 3. Therefore, the delta.percent of TF is 0.064 – 0.809 = -0.745. This allows for communication of the absolute expression level on the y-axis (logFC) and the proportion of cells expressing each gene on the x-axis (delta.percent). This plot is fully interactive, and the user can hover the mouse over each gene to visualize more information regarding the differential expression analysis. Of note, positive logFC and delta.percent values represent increased expression in clusters selected in Group 1. In (D), cells on the dimensionality reduction plot are colored according to their selected group, with cells in group 1 colored red and cells in group 2 colored blue. In (E), differential expression results are displayed in table format. Genes of interest can be reactively searched in the search bar, and the table can be reordered to display genes enriched in each comparison group by arranging genes based on each column. In addition, results from the differential expression can be exported into .csv or .xlsx files.

In contrast, the user may also manually identify groups for differential expression interactively. If the user selects the “manually select populations (draw lasso)” radio button (A), an additional, interactive dimensionality reduction plot appears on the right side of the screen. The lasso tool is automatically selected (arrow, top), and the user can draw a lasso around the cell population for Group 1 on the left plot (B) and for Group 2 on the right plot (C). The user then may select the start differential expression analysis action button, and differential expression results will be performed as displayed in the previous tab.

The user may also perform differential expression between other binary identities corresponding to different biologically meaningful features that are identities in the meta.data. For example, in this experiment, foveal and peripheral libraries were independently prepared. By selecting (A) “Perform differential expression between region,” differential expression is performed between foveal and peripheral libraries for all cells selected in the group (in this case, glial cell populations) (B). This feature is also compatible with the lasso selection tool.