---
title: "scRNA-Seq Standard Pipeline"
canonical: "https://help.biobam.com/space/OED0324/4011098113/scRNA-Seq+Standard+Pipeline"
format: markdown
---
## Pipeline
Figure 1. scRNA-Seq analysis pipeline.
## Data
- Raw sequencing data is available in the NCBI GEO database with accession [GSE122687](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE122687). The FASTQ files can be easily downloaded with the [https://sra-explorer.info/](https://sra-explorer.info/) tool, by specifying the SRA accession number SRP169576.
- 10 samples of *Arabidopsis thaliana* roots from 7 different seedlings were sequenced (named Replicate A to J).
- 3 samples were grown in a high-sucrose medium and 7 in a normal medium.
- Single-cell RNA-Seq libraries were obtained with Drop-seq version 3.1.
- The sequencing was performed with Illumina NextSeq 500 (7 samples), Illumina HiSeq 4000 (2 samples), and Illumina HiSeq 2500 (1 sample).
- Alternatively, download directly the count tables from [this link](https://drive.google.com/file/d/1Hifbq-nKxmyBEkpgDTIkEZxJ6cglHXkj/view?usp=sharing).
- Reference Genome and Annotation: [TAIR10](https://plants.ensembl.org/Arabidopsis_thaliana/Info/Index).
- [Download link](https://ftp.ensemblgenomes.ebi.ac.uk/pub/plants/release-60/fasta/arabidopsis_thaliana/dna/Arabidopsis_thaliana.TAIR10.dna.toplevel.fa.gz) for the genome.
- [Download link](https://ftp.ensemblgenomes.ebi.ac.uk/pub/plants/release-60/gff3/arabidopsis_thaliana/Arabidopsis_thaliana.TAIR10.60.gff3.gz) for the annotation.
## Bioinformatic Analysis
### **1. Quantification**
The first step in the scRNA-Seq analysis pipeline is quantification, that is, determining the expression levels of each gene in each cell within the sample. This process consists of aligning sequencing reads to a reference genome and assigning them to genes based on annotation files (GTF/GFF).
In cell-barcoded library construction technologies (e.g., 10x Chromium, Drop-seq), reads are assigned to individual cells using Cell Barcodes, short sequences added to the read that uniquely tag each cell. Thus, correctly configuring the analysis and specifying the structure of the barcoded reads is crucial for accurate results. This can be easily achieved using the configuration wizard in the [Single Cell RNA-Seq Quantification](https://docs.omicsbox.biobam.com/latest/Single-Cell-RNA-Seq-Quantification/) tool.
To perform the quantification, go to transcriptomics > Single Cell RNA-Seq > Single Cell RNA-Seq Quantification. It is important to note that if the library preparation was performed separately, **different samples may share the same cell barcodes**. Since we want to avoid mixing reads from different samples that have identical barcodes, quantification must be run **separately for each sample**. However, if the library was prepared as a single batch and then split across multiple sequencing lanes, all corresponding FASTQ files should be provided **in the same run** to ensure proper quantification.
For this example dataset, we will run the quantification once per sample.
#### **Parameters**
All parameters remain at their default values, except for the following:
- Reference Genome and GFF/GTF: provide the TAIR10 files.
- Exon Feature: exon.
- Overhang: 100.
- Single-cell RNA-Seq Reads: paired-end mode; provide both FASTQ files for one sample.
- Strandness: Unstranded.
- Preset: Drop-seq.
- UMI Collapsing: UMI Tools.
- UMI Filtering: Multi Mapping UMIs.
- Multimapping Reads: Rescue.
- Feature Counting: Exons+Introns
- Cell Filtering: Empty Drops.
#### **Results**
Each quantification run generates:
- A count matrix** **([Figure 2](#figure2)). It has to be saved for further downstream analysis.
- A summary report ([Figure 3](#figure3)).
- A barcode rank plot ([Figure 4](#figure4)), which helps assess quantification performance.
> Macro (anchor)
Figure 2. Count table double viewer in OmicsBox. The right part shows the cell barcodes and their associated statistics, whereas the left part shows the features and their stats.
> Macro (anchor)
Figure 3. scRNA-Seq quantification summary report.
> Macro (anchor)
Figure 4. UMIs per cell plot for each of the samples.
To evaluate whether the correct number of barcodes (cells) was retained, examine the **Barcode Rank Plot**. This plot shows in the X-axis the cell barcodes, sorted by the total number of UMIs detected, and the Y-axis shows the number of UMIs per barcode. For droplet-based technologies, a **sharp drop** in the plot is typically observed. Beyond this point, detected barcodes likely correspond to background noise, contaminant mRNA, or sequencing errors. The **green-colored barcodes** in the plot represent those retained as actual cells. In this dataset, the number of retained cells is considered **appropriate**, as the green-colored region extends to the middle or just before the sharp drop ([Figure 4](#figure4)).
Additionally, the **alignment rates** ranged from 50% to 80% and the **mean UMIs per cell** was 4,500, ensuring sufficient quantification. This confirms that the quantification process was successful and ready for further analysis.
### **2. Merge Count Matrices**
To analyze all samples together, we first need to merge the count tables. This can be achieved with the [Merge Single Cell Counts](https://docs.omicsbox.biobam.com/latest/Merge-Single-Cell-Counts/) tool (transcriptomics > Single Cell RNA-Seq > Merge Single Cell Counts). This tool does **not** perform data integration; it simply combines all count tables into a single dataset. To prevent duplicate cell barcodes, a **suffix is added to each cell name**, indicating its sample of origin.
#### **Parameters**
- Single Cell Counts: browse for all the count tables to be merged.
- Experimental Design: Specify the experimental design in [Table 1](#) by clicking on the “Add factor” button and typing the values as shown in [Figure 5](#).
Table 1. Experimental Design. > Macro (anchor)
> Macro (anchor)
Figure 5. Experimental design specification in the OmicsBox wizard.
#### **Results**
The tool generates a merged count table containing all cells from all samples ([Figure 6](#)). Each cell name includes a suffix identifying its sample of origin to ensure uniqueness.
All downstream charts and visualizations will now include information from all samples, allowing for a comprehensive analysis.
Figure 6. Merged count table with unique cell barcode IDs.
### **3. Quality Check**
Single-cell RNA-Seq data is known for its **dropout events**, which lead to a high proportion of **zero counts** in the quantification matrix. This makes single-cell data inherently noisy, so it is essential to **filter** out both **cells** with an extremely low number of counts and **genes** that are not expressed in any or only a few cells. Applying these filters increases the specificity of the results and reduces the computational cost of downstream analyses.
Beyond filtering low-count cells and genes, other quality metrics help identify **poor-quality cells**. One is the percentage of counts assigned to **mitochondrial (mtRNA) and/or chloroplast (cpRNA) genes**, which is indicative of cell integrity. A high percentage suggests broken cells, where cytosolic mRNA has been lost, but encapsulated organellar RNA remains intact.
Even though new scRNA-Seq technologies are improving, some scRNA-Seq technologies may generate **doublets or multiplets**, where multiple cells are captured together. These cells typically have a **higher total count of UMIs** and a **greater number of expressed genes** than expected. Removing these artifacts is essential to avoid misleading results, as the combined signal may originate from different cell types.
It is possible to easily detect those cells by identifying outliers in the **distribution of total counts, expressed genes, and mitochondrial gene percentages per cell**, which can be examined using **[Violin Plots](https://docs.omicsbox.biobam.com/latest/Single-cell-RNA-Seq-Count-Matrix/#charts)** (Side Panel > Charts).
#### **Parameters**
For the “% Mitochondrial Genes Distribution” violin plot, provide [this file](https://drive.google.com/file/d/1SkapnglhviptkRXJEJBpRlTHepxwojEa/view?usp=sharing) containing mitochondrial and chloroplastic Arabidopsis thaliana gene IDs.
#### **Results**
The Violin Plot provides an overview of data distribution, while an overlaid Box Plot gives additional statistical details (Q1, median, Q3, whiskers). Outliers typically appear **where the violin shape narrows** or at the **upper/lower whiskers of the box plot** ([Figure 7](#figure7)). The box plot’s whiskers provide a more stringent threshold, while the violin plot threshold is generally recommended for filtering.
For this dataset, the distribution of total counts per cell across all 10 samples shows a narrow upper tail, indicating a small proportion of cells with more than 20,000 counts** **([Figure 8](#figure8)). No lower outliers were detected, as all cells had at least 300 total counts.
The total number of expressed genes per cell is evenly distributed, with only a few outliers at both ends ([Figure 9](#figure9)). Finally, a subset of cells exhibits a high percentage of mitochondrial and chloroplast genes ([Figure 10](#figure10)), suggesting potential broken or dying cells.
These results indicate that there might be some doublets or multiplets and damaged or dying cells in the data. Thus, we should proceed with the filtering.
> Macro (anchor)
Figure 7. Distribution of the total number of counts per cell (dots) in a sample. Both violin and box plots are shown, with orange lines indicating potential outlier thresholds.
> Macro (anchor)
Figure 8. Violin and box plots per sample showing the distribution of total counts per cell (dots).
> Macro (anchor)
Figure 9. Violin and box plots per sample showing the distribution of the total number of expressed genes per cell (dots).
> Macro (anchor)
Figure 10. Violin and box plots per sample showing the distribution of the percentage of mitochondrial and chloroplast genes per cell (dots).
### **4. Filtering**
After identifying outlier cells, the next step is to filter them out to remove potentially misleading data. This can be done using the [Filtering tool](https://docs.omicsbox.biobam.com/latest/Single-cell-RNA-Seq-Filtering/) (Side Panel > Actions > Filtering).
In addition to filtering out low-quality cells, it is also recommended to **filter genes** that are not expressed in a sufficient number of cells, since they typically provide **little biological insight** and can introduce noise into the analysis.
#### **Parameters**
All default settings, except:
- Minimum Cells: 100
- Maximum Counts: 20,000
- Minimum Features: 200
- Filter by % of Mitochondrial Genes: Enabled
- Maximum % Mitochondrial Genes: 10
- Mitochondrial Genes File: [provide this file](https://drive.google.com/file/d/1SkapnglhviptkRXJEJBpRlTHepxwojEa/view?usp=sharing)
#### **Results**
After filtering, the resulting count table contains approximately 1,000 fewer cells and 13,000 fewer genes compared to the original dataset. This filtering step will reduce noise in downstream analyses and improve their computational efficiency. To confirm that filtering was applied correctly, the same charts as in the previous section can be regenerated for verification.
### **5. Clustering**
Single-cell RNA-Seq clustering aims to identify groups of cells with **similar expression patterns**. These clusters serve as the foundation for downstream analyses, since, theoretically, cells with similar expression patterns should belong to the same cell type, step, or have condition-specific differences.
Clustering is often the first major insight in a single-cell RNA-Seq analysis. However, raw counts, even after filtering, cannot be directly used for clustering. Thus, a typical scRNA-Seq pipeline involves several **preprocessing steps**, including normalization, dimensionality reduction, and data integration. All these preprocessing steps, along with clustering, can be performed using [OmicsBox’s Seurat implementation](https://docs.omicsbox.biobam.com/latest/Single-Cell-RNA-Seq-Clustering/) (Side Panel > Actions > Clustering).
#### **Parameters**
All default settings, except:
- Data correction > Mitochondrial Genes File: Enabled and provide[ this file](https://drive.google.com/file/d/1SkapnglhviptkRXJEJBpRlTHepxwojEa/view?usp=drive_link)
- Integration Factor: medium
- Integration Method: Seurat-CCA
- N. Dimensions for Integration: 35
- Define Dimensions by: Manual
- Number of Dimensions: 20
- Resolution: 0.6
#### **Results**
The clustering analysis identified a total of **22 clusters** ([Figure 11](#figure11)), incorporating cells from all growth media ([Figure 12](#figure12)) and all samples ([Figure 13](#figure13)). This indicates that **integration was successful**, effectively grouping cells with similar expression patterns across both conditions. Notably, Clusters 11, 14, and 18 contain a higher proportion of cells grown in high-sucrose medium, suggesting a potential condition-dependent effect on cell distribution.
> Macro (anchor)
Figure 11. UMAP colored by assigned cluster.
> Macro (anchor)
Figure 12. UMAP colored by growing medium.
> Macro (anchor)
Figure 13. UMAP colored by the sample of origin.
### **6. Cell Type Prediction**
Cell type prediction identifies the cell types present in the dataset based on prior knowledge. This is a **critical** step in the pipeline, as all subsequent biological insights rely on accurate cell-type annotation and will be based on the predicted cell type.
Common reference-based approaches involve comparing gene expression patterns between our query dataset and a pre-annotated single-cell RNA-Seq reference. The cell type is then assigned based on gene expression similarity. Thus, choosing a **suitable reference** is crucial for ensuring **accurate predictions**.
For this dataset, we will use a root *Arabidopsis thaliana* reference from the [scPlantDB database](https://biobigdata.nju.edu.cn/scplantdb/dataset), which includes 200,000 cells from 19 samples and 16 distinct cell types. The cell-type prediction will be performed using OmicsBox’s [SingleR implementation](https://docs.omicsbox.biobam.com/latest/Single-Cell-RNA-Seq-Cell-Type-Prediction-SingleR/) (Side Panel > Actions > Cell Type Prediction > SingleR).
#### **Parameters**
All by default except:
- Reference Format: H5 Annotated Data
- Reference: browse for the [reference annotation](https://drive.google.com/file/d/10yxonA8QjB5DeNpsM4YJGNTrjV7q_yPz/view?usp=drive_link)
- Cell Types: Celltype
- Select Query Feature: ID
- Select Matching Reference Feature: features
- Gene Marker Selection Method: Classic
#### **Results**
[Figure 14](#figure14) presents the UMAP projection with cells colored by predicted cell types. A total of **16 different cell types** were identified, along with an additional “Pruned” category, which includes cells with unreliable annotations. Interestingly, most pruned cells are grouped within Cluster 17 ([Figure 11](#figure11)), suggesting that this cluster likely consists of low-quality cells with misleading data or a distinct cell type that is not present in the reference dataset
To further evaluate the **specificity of the predictions**, we can examine the [Score Heatmap](https://docs.omicsbox.biobam.com/latest/Single-Cell-RNA-Seq-Cell-Type-Prediction-SingleR/#scores-heatmap) ([Figure 15](#figure15)). In an ideal scenario, cells should have high prediction scores for only one cell type—as seen for “Root Hair” in [Figure 15](#figure15). However, fuzzy areas in the right section of the heatmap indicate cells with multiple candidate cell types.
To further validate the specificity of predictions, we can refer to the [Delta Score Distribution](https://docs.omicsbox.biobam.com/latest/Single-Cell-RNA-Seq-Cell-Type-Prediction-SingleR/#delta-distribution-chart) plot ([Figure 16](#figure16)). This plot measures the difference between the assigned label score and the median score across all labels. Ideal predictions should have high delta scores and few pruned cells (marked in yellow), as these are cells that could not be confidently assigned to a specific type.
Overall, the annotation quality is considered good, as the specificity is high, with fuzzy areas limited to similar cell types, there are few pruned labels, and cell-type annotations align well with the identified clusters in the UMAP, meaning annotations are not randomly dispersed.
> Macro (anchor)
Figure 14. UMAP projection with cells colored by the cell type predicted by SingleR with the scPlantDB reference SRP267870.
> Macro (anchor)
Figure 15. SingleR score heatmap with cells in columns and labels in rows. The value is the score given for a particular cell-label pair. The labels displayed on the top legend are the final label assignment.
> Macro (anchor)
Figure 16. SingleR delta Score distribution for each predicted cell type. Cells colored in yellow have been pruned.
### **7. Performing Custom Annotations**
By leveraging **clustering results**, **UMAP dimensional reduction**, and **cell-type annotations**, we can manually refine cell groupings using the [tools available in the UMAP Side Panel](https://docs.omicsbox.biobam.com/latest/Single-Cell-UMAP-tSNE-Viewer/). This allows for **custom cell classifications** beyond those provided by the automated annotation process.
#### **Using the Selection Tools**
Cluster-17 ([Figure 11](#figure11)) consists primarily of cells labeled as “Pruned” in the cell type prediction ([Figure 14](#figure14)), suggesting that these may either be poor-quality cells grouped together or a cell type not included in the reference annotation. To further investigate them, we will temporarily assign these cells to a group named “Unknown” following these steps ([Figure 17](#figure17)):
1. Go to the UMAP Side Panel > Selection.
1.1. Click on the Lasso tool, hold the left mouse button, and circle the target cells.
2. Go to the UMAP Side Panel > Assign Group to Selection.
2.1. In the “Group” text box, enter a name (e.g. “My Annotation”).
2.2. In the “Value” text box, specify a new cell type (e.g. “Unknown”).
2.3. Click Apply.
> Macro (anchor)
Figure 17. Example of lasso selection tool in the UMAP/tSNE viewer and assignment of selected cells to a new custom group.
#### **Using the Advanced Selection Tool**
In some cases, we may need to group cells that share specific categorical attributes. For instance, cells labeled as “G2/M Phase” and “S Phase” in the prediction results ([Figure 14](#figure14)) are predominantly found in Cluster-4 ([Figure 11](#figure11)). These cells likely represent actively dividing cells, so we might want to group them under a broader category, such as "Dividing Cells". To achieve so, we have to follow these steps ([Figure 18](#figure18)):
1. Go to UMAP Side Panel > Advanced Selection.
1.1. Click on “Add” in the “Categorical Rules” and set the dropdowns to “in”, “seurat_clusters”, and “Cluster-4”.
1.2. Click on “Apply”.
2. Go to the UMAP Side Panel > Assign Group to Selection.
2.1. Select the previously created group in the “Group” dropdown.
2.2. Type a new cell type in the “Value” text box, e.g. “Dividing Cells”.
2.3. Click on “Apply”.
> Macro (anchor)
Figure 18. Example of selection of cells by a categorical rule in the UMAP/tSNE viewer and assignment of selected cells to a preexisting custom group.
### **8. Differential Expression Analysis**
Once cell groups are defined—whether based on clusters or cell types—the dataset can be analyzed for differential gene expression. For example, we can investigate whether gene expression differs between growing conditions (high sucrose vs. normal medium) within each cell type.
This can be achieved using the [EdgeR implementation in OmicsBox](https://docs.omicsbox.biobam.com/latest/Single-Cell-RNA-Seq-Differential-Expression-Analysis/) (Side Panel > Actions > Differential Expression), which is specifically adapted for single-cell RNA-Seq data.
#### **Parameters**
To compare gene expression between cells grown in normal vs. sucrose medium for each cell type, configure the analysis as follows:
- CPM Filter: 2
- Cells Reaching CPM Filter: 200
- Multifactorial Design
- Biological Replicates: Sample
- Primary Factor: athaliana_root_SRP267870_pruned
- Primary Contrast Conditions: select all except “Pruned”
- Test Contrasts Separately: Checked
- Secondary Factor: Medium
- Secondary Contrast Condition: Sucrose
- Secondary Reference Condition: Normal
#### **Results**
The main output is a differential expression table containing gene statistics for each test performed ([Figure 17](#figure17)). Additionally, each gene is tagged as UP or DOWN if it is significantly upregulated or downregulated, respectively. The thresholds used to define the tags can be adjusted using Side Panel > Actions > Set UP/DOWN Tags. A summary table is also generated, displaying the number of differentially expressed genes (DEGs) per test.
Key findings for this dataset are that most cell types show few or no DEGs between conditions. However, some cell types (e.g., Non-hair and Root Cortex) exhibit **notable differences** in gene expression between normal and high sucrose conditions.
> Macro (anchor)
Figure 17. Differential Expression Analysis of cells grown in high-sucrose medium versus normal cells for each cell type.
### **9. Functional Enrichment**
Interpreting the list of differentially expressed genes (DEGs) can be challenging, so to simplify this task, we can perform a [Fisher’s Exact Test](https://docs.omicsbox.biobam.com/latest/Single-Cell-RNA-Seq-Differential-Expression-Analysis/#fishers-exact-test) (Side Panel > Actions > Fisher’s Exact Test) to **identify biological functions** that are significantly enriched. This helps provide a more meaningful interpretation of the results.
#### **Parameters**
All default settings, except:
- Test-Set genes: Up-regulated genes
- Contrast to Test: “Root cortex.Sucrose”
- Reference Annotation: provide[ this functional annotation file](https://drive.google.com/file/d/1mBxI5QXbfPUq9G_2k38xJjEQ5_65VlLQ/view?usp=drive_link)
#### **Results**
The main output is a table listing GO terms along with statistical values, including an OVER or UNDER tag for terms found to be over- or under-represented ([Figure 18](#figure18)). At first glance, many enriched functions related to **stress response** can be observed, suggesting that high sucrose levels in the medium may significantly affect these cell types.
To better visualize these results, several charting options are available:
- Enriched GO Graph (Side Panel > Charts > Enriched GO Graph): Displays the hierarchical structure of enriched GO terms (Figure 19).
- Word Cloud** **(Available from the Enriched GO Graph Side Panel): Provides an intuitive visualization of the most frequently enriched functions (Figure 20).
- Bubble Plot (Side Panel > Charts > Bubble Plot): Helps assess the significance of enrichment results. To match the view in Figure 21, adjust the chart's Side Panel settings:
- Max Bubble Size: 50
- Manual Axis Limits (X-Axis Options):
- Min Value: 0
- Max Value: 0.5
> Macro (anchor)
Figure 18. Fisher’s Exact Test main table results.
> Macro (anchor)
Figure 19. Enriched GO Graph of the category Biological Process. Each box is colored according to the number of sequences annotated with that GO.
> Macro (anchor)
Figure 20. Word Cloud with the names of the enriched GO terms.
> Macro (anchor)
Figure 21. Bubble plot of the enriched GO terms. The X-axis shows the gene ratio (Nr Test / [Nr Test + Not Annot Test]), the dot size is the number of test sequences, and the color is the adjusted p-value (darker color means lower p-values).
## **Conclusions**
Overall, these findings provide evidence that high-sucrose mediums induce a cell-type-specific response, as differences in gene expression were not observed across all cell types. This highlights the power of single-cell RNA-Seq, which enables the identification of **specific cellular behaviors** that would be masked in traditional bulk RNA-Seq analyses. A bulk approach would have likely detected a generalized stress response, but it would not have revealed the specific effects on individual cell types.
By leveraging single-cell transcriptomics, we can gain a more precise and comprehensive understanding of how different cell populations react to environmental changes, paving the way for more targeted biological insights and experimental validations.