In compbiomed/singleCellTK: Comprehensive and Interactive Analysis of Single Cell RNA-Seq Data
````{=html}
## Introduction
Celda is a Bayesian hierarchical model that can perform bi-clustering of features into modules and observations into subpopulations.
To view instructions on how to apply Celda to single-cell RNA sequencing (scRNA-seq) datasets, please select 'Interactive Analysis' for using Celda in shiny application or 'Console Analysis' for using this package on R console from the tabs below. The detailed tutorial of Celda can be found at [Celda's official website](https://www.camplab.net/celda/).
## Workflow Guide
````{=html}
<div class="tab">
<button class="tablinks" onclick="openTab(event, 'interactive')" id="ia-button">Interactive Analysis</button>
<button class="tablinks" onclick="openTab(event, 'console')" id="console-button">Console Analysis</button>
</div>
<div id="interactive" class="tabcontent">
Entry of The Panel
In this tutorial example, we illustrate all the steps of the curated workflow and focus on the options available to manipulate and customize the steps of the workflow as per user requirements. From anywhere of the UI, the panel for celda can be accessed from the top navigation panel at the circled tab shown below by simply clicking on the 'Curated Workflows' from the top menu and then select Celda:
\
NOTE: This tutorial assumes that the data has already been uploaded via the upload tab of the toolkit and filtered before using the workflow.
Run Celda Analysis
1. Identify number of feature modules
To identify number of feature modules, there are always five essential inputs that users should be sure with:
- The data matrix to use - selection input "Choose an Assay". In terms of celda analysis, SCTK always requires a full-sized feature expression data (i.e. assay) as a valid input.
- Feature selection method - selection input "Choose Feature Selection Method". Select feature selection method from 'None' , 'runSeuratFindHVG' and 'Scran_modelGeneVar'.
- Minimum number of counts required for feature selection - integer "Keep features with this many counts".
- Minimum number of cells required for feature selection - integer "In at least this many cells".
- Number of initial feature modules - integer "Select Number of Initial Feature Modules".
- Number of maximum feature modules - integer "Select Number of Maximum Feature Modules".
\
After the feature selection method is confirmed, the lower part will dynamically switch to the method specific settings.
````{=html}
runSeuratFindHVG
The method specific parameters for `runSeuratFindHVG` includes:
- Select method for computation of highly variable genes - selection input **"Select HVG method"**. `Seurat` provides three methods for variable genes identification i.e. `vst` (uses local polynomial regression to fit a relationship between log of variance and log of mean), `mean.var.plot` (uses mean and dispersion to divide features into bins) and `dispersion` (uses highest dispersion values only).
- Select variable features - selection input **"Select number of variable features"**.Input the number of genes that should be identified as variable. Default is 2500.
````{=html}
</details>
<details>
<summary><b>Scran_modelGeneVar</b></summary>
The method specific parameters for Scran_modelGeneVar includes:
- Select variable features - selection input "Select number of variable features".Input the number of genes that should be identified as variable. Default is 2500.
````{=html}
After feature selection method and it's parameters are set, user can click recursive module split button to run `recursiveSplitModule` function which is used used to cluster features into modules for a range of feature modules (L). Two plots are shown when user runs recursive module split; perplexity plot and rate of perplexity change.
To evaluate the performance of individual models we use perplexity plot by calculating the probability of observing expression counts given an estimated Celda model. The perplexity alone often does not show a clear elbow or “leveling off”. However, the rate of perplexity change (RPC) can be more informative.
\
\
**2. Identify number of cell clusters**
To identify number of cell clusters, there are always two essential inputs that users should be sure with:
- Number of initial cell modules - integer **"Select Number of Initial Cell Modules"**. Initial number of cell populations to try. Default 5.
- Number of maximum cell modules - integer **"Select Number of Maximum Cell Modules"**. Maximum number of cell populations to try. Default 25.
\
After these two parameters are set, user can click recursive cell split button to run `recursiveSplitCell` function to cluster cells into population for range of possible sub populations(K). Three plots are shown when user runs recursive cell split; perplexity plot, rate of perplexity change and preliminiary UMAP plot.
To evaluate the performance of individual models we use perplexity plot by calculating the probability of observing expression counts given an estimated Celda model. The perplexity alone often does not show a clear elbow or “leveling off”. However, the rate of perplexity change (RPC) can be more informative.
\
\
UMAP plot is useful for visualizing the relationship between cells.
\
**3. Visualization**
After selecting a celda model with specific values of L and K, we can then perform additional exploratory and downstream analyses to understand the biology of the transcriptional modules and cell populations.
````{=html}
<style>
div.offset { background-color:inherit ; border-radius: 5px; margin-left: 15px;}
</style>
<div class = "offset">
<details>
<summary><b>Embeddings</b></summary>
To visualize the relationships between the cells in a 2-D embedding we can generate a dimension reduction plot with the Uniform Manifold Approximation and Projection (UMAP) or a t-distributed stochastic neighbor embedding (t-SNE) method. User can select from UMAP and TSNE that which plot they want to visualize.
\
````{=html}
Module Analysis
The Module Analysis tab contains the probabilities and heatmaps for each module. Module probability plots will color cells by the probability of each module on a 2-D embedding plot. Module heatmaps show the relative expression for the features in a module across the cells. Cells within the heatmap will be ordered from the lowest to highest probability of the module. The parameter topCells can be used to control the number of cells included in the heatmap.
\
````{=html}
</details>
<details>
<summary><b>Probability Map</b></summary>
Celda has the ability to identify modules of co-expressed features and quantify the probability of these modules in each cell population. An overview of the relationships between modules and cell subpopulations can be explored with the function celdaProbabilityMap.The “Absolute probability” map gives insights into the absolute abundance of a module within a given cell subpopulation. The absolute heatmap can be used to explore which modules are higher than other modules within a cell population. The “Relative expression” map shows the standardized (z-scored) module probabilities across cell subpopulations. The relative heatmap can be used to explore which modules are relatively higher than other modules across cell populations.
\
````{=html}
**4. Downstream Analysis**
Once all steps of the celda workflow are completed, users can further analyze the data by directly going to the various downstream analysis options (Differential Expression, Marker Selection & Pathway Analysis) from within the celda workflow.
\
````{=html}
<div class = "offset">
````{=html}
## Introduction
Celda is a Bayesian hierarchical model that can perform bi-clustering of features into modules and observations into subpopulations.
To view instructions on how to apply Celda to single-cell RNA sequencing (scRNA-seq) datasets, please select 'Interactive Analysis' for using Celda in shiny application or 'Console Analysis' for using this package on R console from the tabs below. The detailed tutorial of Celda can be found at [Celda's official website](https://www.camplab.net/celda/).
## Workflow Guide
````{=html}
<div class="tab">
<button class="tablinks" onclick="openTab(event, 'interactive')" id="ia-button">Interactive Analysis</button>
<button class="tablinks" onclick="openTab(event, 'console')" id="console-button">Console Analysis</button>
</div>
<div id="interactive" class="tabcontent">
Entry of The Panel
In this tutorial example, we illustrate all the steps of the curated workflow and focus on the options available to manipulate and customize the steps of the workflow as per user requirements. From anywhere of the UI, the panel for celda can be accessed from the top navigation panel at the circled tab shown below by simply clicking on the 'Curated Workflows' from the top menu and then select Celda:
\
NOTE: This tutorial assumes that the data has already been uploaded via the upload tab of the toolkit and filtered before using the workflow.
Run Celda Analysis
1. Identify number of feature modules
To identify number of feature modules, there are always five essential inputs that users should be sure with:
- The data matrix to use - selection input "Choose an Assay". In terms of celda analysis, SCTK always requires a full-sized feature expression data (i.e. assay) as a valid input.
- Feature selection method - selection input "Choose Feature Selection Method". Select feature selection method from 'None' , 'runSeuratFindHVG' and 'Scran_modelGeneVar'.
- Minimum number of counts required for feature selection - integer "Keep features with this many counts".
- Minimum number of cells required for feature selection - integer "In at least this many cells".
- Number of initial feature modules - integer "Select Number of Initial Feature Modules".
- Number of maximum feature modules - integer "Select Number of Maximum Feature Modules".
\
After the feature selection method is confirmed, the lower part will dynamically switch to the method specific settings.
````{=html}
runSeuratFindHVG
The method specific parameters for `runSeuratFindHVG` includes:
- Select method for computation of highly variable genes - selection input **"Select HVG method"**. `Seurat` provides three methods for variable genes identification i.e. `vst` (uses local polynomial regression to fit a relationship between log of variance and log of mean), `mean.var.plot` (uses mean and dispersion to divide features into bins) and `dispersion` (uses highest dispersion values only).
- Select variable features - selection input **"Select number of variable features"**.Input the number of genes that should be identified as variable. Default is 2500.
````{=html}
</details>
<details>
<summary><b>Scran_modelGeneVar</b></summary>
The method specific parameters for Scran_modelGeneVar includes:
- Select variable features - selection input "Select number of variable features".Input the number of genes that should be identified as variable. Default is 2500.
````{=html}
After feature selection method and it's parameters are set, user can click recursive module split button to run `recursiveSplitModule` function which is used used to cluster features into modules for a range of feature modules (L). Two plots are shown when user runs recursive module split; perplexity plot and rate of perplexity change. To evaluate the performance of individual models we use perplexity plot by calculating the probability of observing expression counts given an estimated Celda model. The perplexity alone often does not show a clear elbow or “leveling off”. However, the rate of perplexity change (RPC) can be more informative. \ \ **2. Identify number of cell clusters** To identify number of cell clusters, there are always two essential inputs that users should be sure with: - Number of initial cell modules - integer **"Select Number of Initial Cell Modules"**. Initial number of cell populations to try. Default 5. - Number of maximum cell modules - integer **"Select Number of Maximum Cell Modules"**. Maximum number of cell populations to try. Default 25. \ After these two parameters are set, user can click recursive cell split button to run `recursiveSplitCell` function to cluster cells into population for range of possible sub populations(K). Three plots are shown when user runs recursive cell split; perplexity plot, rate of perplexity change and preliminiary UMAP plot. To evaluate the performance of individual models we use perplexity plot by calculating the probability of observing expression counts given an estimated Celda model. The perplexity alone often does not show a clear elbow or “leveling off”. However, the rate of perplexity change (RPC) can be more informative. \ \ UMAP plot is useful for visualizing the relationship between cells. \ **3. Visualization** After selecting a celda model with specific values of L and K, we can then perform additional exploratory and downstream analyses to understand the biology of the transcriptional modules and cell populations. ````{=html} <style> div.offset { background-color:inherit ; border-radius: 5px; margin-left: 15px;} </style> <div class = "offset"> <details> <summary><b>Embeddings</b></summary>
To visualize the relationships between the cells in a 2-D embedding we can generate a dimension reduction plot with the Uniform Manifold Approximation and Projection (UMAP) or a t-distributed stochastic neighbor embedding (t-SNE) method. User can select from UMAP and TSNE that which plot they want to visualize.
\
````{=html}
Module Analysis
The Module Analysis tab contains the probabilities and heatmaps for each module. Module probability plots will color cells by the probability of each module on a 2-D embedding plot. Module heatmaps show the relative expression for the features in a module across the cells. Cells within the heatmap will be ordered from the lowest to highest probability of the module. The parameter topCells can be used to control the number of cells included in the heatmap.
\
````{=html}
</details>
<details>
<summary><b>Probability Map</b></summary>
Celda has the ability to identify modules of co-expressed features and quantify the probability of these modules in each cell population. An overview of the relationships between modules and cell subpopulations can be explored with the function celdaProbabilityMap.The “Absolute probability” map gives insights into the absolute abundance of a module within a given cell subpopulation. The absolute heatmap can be used to explore which modules are higher than other modules within a cell population. The “Relative expression” map shows the standardized (z-scored) module probabilities across cell subpopulations. The relative heatmap can be used to explore which modules are relatively higher than other modules across cell populations.
\
````{=html}
**4. Downstream Analysis**
Once all steps of the celda workflow are completed, users can further analyze the data by directly going to the various downstream analysis options (Differential Expression, Marker Selection & Pathway Analysis) from within the celda workflow.
\
````{=html}
<div class = "offset">
````{=html}
