Introduction
Welcome to shinyEscort, a user-friendly Shiny application designed to help researchers in constructing more accurate trajectories on their single-cell RNA-seq data. With shinyEscort, you input your original data and normalized data, and the application walks you through our evaluation framework.
Getting started
In R, assuming you have successfully downloaded Escort, then you can start the Shiny app locally by running the following:
library(Escort)
library(shiny)
shinyEscort()If you encounter any issues, please report these to our Github: Report Issues
About
An overview of shinyEscort and the Escort workflow is provided in the “About” tab positioned at the top of the left column sidebar.
Step 1
Input:
To initiate the analysis assessing a dataset’s evidence of a trajectory signal, click on the “Step 1” button in the left column sidebar. Input must include raw and normalized single-cell RNA-seq datasets in .csv format. Then click the “Import” button. Data should have undergone basic quality control removing low quality cells, doublets, and empty droplets.
For hands-on practice, example datasets are stored in the data directory of the R package featuring simulated scenarios such as homogeneous data, datasets with distinct cluster types, and linear topology datasets. In this demonstration, we utilize “rawcount_linear_splatter.csv” and “normcount_linear_splatter.csv” which showcases a simulated scRNA-seq data with a linear structure.
Output:
The number of cells and genes
Internally,
Escort checks the compatibility of normalized and raw dataset to ensure
they originate from the same source. If they are consistent, the number
of genes and cells will be displayed. In cases where the datasets do not
align or if there is any missing input data, an “NA” will be displayed
in the number of genes and cells. Users should re-examine whether they
input the correct datasets.
UMAP and t-SNE representations
On the right
side, UMAP and t-SNE representations provide a visual representation of
the datasets.
Existence of trajectory
Two scenarios
considered inappropriate for trajectory inference include cells from
biologically distinct clusters and sets of cells with insufficient
heterogeneity. These scenarios are evaluated in the result boxes labeled
“Distinct cell types” and “Homogeneous cells.” The determination of
trajectory existence is displayed above the UMAP and t-SNE
representations.
Diverse cell types In instances where the dataset fails this evaluation, detecting distinct clusters triggers a subsequent differential expression analysis utilizing the edgeR R package indicating cluster-specific genes. We display the top 30 differentially expressed genes between pairwise clusters. Users should examine whether fitting a trajectory that connects these cell types is biologically reasonable. If so, then users should re-examine whether intermediate cell types exist, presence of batch effects, and choice of normalization method.
Homogeneous cells For datasets demonstrating too little heterogeneity, a list of highly variable genes is presented in the “HVGs” subtab, followed by a Gene Ontology (GO) enrichment analysis, and the results are presented within the “GO” subtab. Again, the appropriateness of an underlying trajectory should be considered. To proceed, users should investigate whether other processes could be overriding the biological signal of interest (e.g. cell cycle) or if excessive signal from ribosomal or mitochondrial gene is present. Any upstream pre-processing may also need to be re-examined or eliminated.
Generate embeddings
Input:
Custom embeddings for evaluation can be generated in this tab. For now, we offer the selection of a dimension reduction technique (UMAP, MDS, PCA, or TSNE) and the number of highly variable genes. By default, Escort utilizes Slingshot for the preliminary trajectory fitting that takes place in Step 3. Users may also generate their own embedding options and skip this tab. How do to so is described in the Generating data objects for Escort R/Shiny Vignette.
Output:
Visualization
Visualize the embedding and
trajectory on the right.
Download .rds
Click “Download” to save the
embedding as an .rds file that can be uploaded in the next step.
Step 2:
Input:
Next, in Step 2, users upload any .rds files generated in the prior step (or generated on their own). In this demonstration, we upload six .rds files that were generated, as shown in the accompanying screenshot.
Output:
Escort evaluates embeddings based on inter-cellular relationships,
preservation of similarity relationships, and cell density. Results are
presented in three tables.
Inter-cellular relationships
Evaluates cell
connectivity on the embedding. A result of “TRUE” followed by a
checkmark “✓” indicates the absence of disconnected clusters in the
embedding.
Preservation of similarity relationships
Determines the preservation score for similarity relationships,
reflecting the percentage of cells exhibiting a high level of
similarity. A higher value indicates a greater number of cells
performing well in preserving similarity relationships.
Cell density
Examines the distribution of
cells in the embedding. A higher value indicates a more uniform
distribution in the two-dimensional embedding space. This uniformity
poses challenges for trajectory inference methods in identifying a
robust trajectory.
Step 3
Input:
No manual input required; calculations are performed automatically once Step 2 begins.
Output:
Utilizing the embedding-specific preliminary trajectory, Escort
estimates the proportion of cells with ambiguous projections along the
trajectory, displaying the results in a table. A higher value indicates
more ambiguous cells in the embeddings. 
Summary
Input:
No manual input required; calculations are performed automatically once Step 2 begins.
Output:
Evaluation Table
Escort uses a standardized
benchmark scoring system ranging from negative infinity to two. Higher
scores indicate better performance. Embeddings with a score greater than
zero are recommended, while those with a score less than or equal to
zero are considered non-recommended. Users can easily identify optimal
embedding choices for constructing a trajectory through this scoring
system. The conclusive results, featuring embeddings and trajectories,
are presented in a table sorted by score.
Representations
A visual representation of the
top 6 highest-scoring embeddings and trajectories is shown below the
table. 