Lecture 9 Single-Cell RNA Sequencing

Lecture 9 - Single Cell RNA-Seq

• Single-cell RNA sequencing (scRNA-seq) is a high-resolution method that analyzes gene expression at the level of individual cells. It captures cellular heterogeneity and uncovers subpopulations within tissues.
• ScRNA-seq is useful for complex tissues with diverse cell types, helps identify individual cell differences, isolates each cell's transcriptome, and uncovers variations in rare cell types.
• Applications include developmental biology, cancer research, immunology, neuroscience, and signal transduction.
• Key steps in scRNA-seq data processing move from quality control to normalization.
• Differential gene expression (DGE) analysis is used to identify the expression of genes in single-cell data.
• Cell lineage and RNA velocity techniques provide dynamic insights into cellular differentiation and development.
• Pathway and functional enrichment analysis helps interpret the biological significance of identified cell clusters.

Where are we going?

• The workflow in single cell RNA sequencing proceeds through DNA sequencing, sequencing quality control, DNA assembly, DNA read mapping, genome annotation, and expression analysis. Other sections of the pathway are marker-trait associations, population analysis, and genotyping that look at polymorphisms.

Learning Outcomes

• Students will describe the basics of scRNA-seq molecular biology including how gene expression is captured at the single-cell level.
• Students will outline the steps in scRNA-seq data processing, from quality control to normalization.
• Students will identify methods used for differential gene expression (DGE) analysis in single-cell data.
• Students will explain how cell lineage and RNA velocity techniques provide dynamic insights into cellular differentiation and development.
• Students will apply pathway and functional enrichment analysis to interpret biological significance in identified cell clusters.

Introduction to scRNA-seq

• scRNA-seq is a high-resolution method for analyzing gene expression in individual cells.
• Important because it captures cellular heterogeneity and reveals subpopulations.
• Applications are widespread, including developmental biology, cancer research, immunology, neuroscience, and signal transduction.

Why Single-Cell Analysis?

• Bulk RNA sequencing averages signals across cells, potentially masking individual cell differences.
• scRNA-seq isolates each cell's transcriptome, uncovering variations and rare cell types, useful for complex tissues with diverse cell types.

Applications of scRNA-seq in Research

• Cell type identification reveals new cell types and biomarkers.
• Tissue heterogeneity shows which rare cell types can have a big impact on health and disease.
• Drug target discovery helps uncover new drug targets.
• Cell development pathways can be reconstructed with scRNA-seq.
• Immune profiling is used with scRNA-seq.
• Cancer profiling maps and analyzes CNVs (copy number variations) in cancer.

Single Cell Library Platforms

• 10X Genomics Chromium is the most popular platform for creating sequencing libraries from single cells.
• Other platforms have throughput ranging from low to high, and sensitivity and data quality vary.
• 10x Genomics Chromium uses microfluidics to encapsulate individual cells in droplets. Also employs a 3'-tag sequencing method for capturing the 3' end of each mRNA transcript.

10X Genomics Chromium Prep

• The image shows a visual representation of isolating cells, preparing them, and running the cDNA process with the 10x Chromium. The different steps involved are indicated in the figure.

scRNA-Seq Data Processing

• Data processing includes steps such as BCL file processing, signal processing, sequencing reads, QC of FASTQ, alignment, Spliced alignment to genome, Lightweight mapping to the (extended) txome, count assignment, UMI resolution, CB correction, and quantification.

Sequencing Data QC

• Ensures data accuracy and reliability.
• Identifies potential issues in early analysis, reducing errors in downstream analysis.
• Quality Metrics include Base Quality Scores (sequencing accuracy), Read Length Distribution (consistency and length), and Adapter Sequence Detection (identifying adapter sequences).

Key QC Metrics

• High base quality scores indicate reliable base calls, while quality usually declines towards the end of a read.
• Uniform read length across sequences is ideal.
• Shorter or variable read lengths can indicate sequencing issues or degradation.
• Adapters should be removed before downstream analysis with tools like Trimmomatic or Cutadapt.

Read Mapping in scRNA-Seq

• Mapping aligns sequencing reads to a reference to identify gene expression levels.
• Types of Mapping include Genome Mapping (aligns reads to the entire genome to identify potential sequences and splicing events), Transcriptome Mapping (aligns reads to known transcripts which potentially misses novel transcripts), and Augmented Transcriptome Mapping (aligns reads to known transcripts plus splicing events and balances speed and accuracy for complex analysis).

10X Genomics Read Structure

• The figure shows the structure of the sequenced reads from a 10X Genomics pipeline, that is the order of read 1, barcode, UMI, and poly(dT)VN, etc.

Cell Barcode Correction

• Barcodes are unique DNA sequences used to identify reads from individual cells in scRNA-seq experiments.
• Errors in barcodes can misassign reads to the wrong cells, requiring correction through methods like Hamming Distance Correction and Cluster-Based Correction.
• Sequencing Errors (random errors in base calling leading to mismatched nucleotides) and Synthesis Errors (Errors during barcode synthesis) are two types of errors.

Methods for Cell Barcode Correction

• Algorithmic approaches use Hamming Distance Correction to calculate differences between barcodes, correcting errors with only 1-2 differences.
• Cluster-Based Correction groups similar barcodes, assigning them to the most probable correct sequence within a cluster.
• Filtering Techniques like Ambiguous Barcode Filtering can exclude low-quality barcodes.
• Consensus-Based Correction predicts the likely original sequence based on data patterns, especially useful in high-throughput systems.

Challenges with Cell Barcode Correction

• Distinguishing true biological diversity from technical errors in barcodes is a challenge.
• Over-correction may mistake a unique cell's barcodes.
• Large datasets may have high levels of barcode noise, complicating correction.
• Sophisticated corrections may require significant computational resources.
• Machine Learning integration and improved error models can offer improved accuracy.

Unique Molecular Identifiers

• UMIs are short, random sequences added to each mRNA molecule before PCR amplification.
• They uniquely identify each transcript, distinguishing uniquely sequenced transcripts from PCR duplicates.
• UMIs are important because they reduce amplification bias, distinguishing unique transcripts from duplicated ones and lead to accurate gene expression quantification.

Graph-Based UMI Resolution

• UMIs are represented as nodes in a graph based on their similarity (differing by one base).
• Connected nodes represent likely duplicates, resolved through clustering.
• The number of unique UMIs per gene is counted to accurately estimate transcript abundance.

Challenges of UMI Resolution

• Misreads in UMI sequences can introduce errors that are difficult to distinguish from true duplicates,
• Graph-based methods can be computationally intensive, requiring careful tuning.

Empty Droplet Removal

• Empty droplets (droplets with no cells) can still capture environmental RNA, leading to background noise.
• Removing empty droplets is crucial for accurate expression profiles.
• Further quality control of the data is necessary after UMIs have been identified.

Strategies for Empty Droplet Removal

• Threshold-based filtering sets a minimum threshold for transcripts per droplet, excluding those below.
• Ambient RNA profiling identifies characteristic gene expression patterns to mark droplets for removal.
• Statistical methods, like EmptyDrops, uses models to differentiate real cells from empty droplets based on transcript distribution.

Double Detection

• Doublets occur when two or more cells are captured in a single droplet.
• This can lead to mixed gene expression profiles and inaccurate data if not detected.
• Doublet detection is essential because doublets can create artificial cell types or clusters, which affects downstream analyses.

Doublet Removal

• Density-based clustering flags cells with unusually high gene or UMI counts.
• Gene expression patterns of mixed profiles can identify doublets.
• Tools, like Scrublet and DoubletFinder, identify doublets based on expected cell-to-cell gene expression similarity.

Count Data Normalization

• Normalizing count data reduces noise biases to meaningfully compare cells and conditions.
• Raw counts are direct counts of RNA transcripts per gene in each cell after processing.
• Normalized counts adjust the counts to account for differences in sequencing depth or cell size, using methods like CPM (Counts Per Million) or TPM (Transcripts Per Million).

Count Data Normalization (Log-transformed, Scaled)

• Logarithmic transformation of normalized counts stabilize variance across genes.
• Scaled counts standardize counts (centering and scaling) useful for dimensionality reduction techniques.

Variance Stabilization

• Some analysis methods do not prefer data that results from variance standardization.

Overview of scRNA-Seq Analysis

• The goal of scRNA-Seq analysis is to identify patterns in gene expression across individual cells, discovering unique cell types, functional states, and biological pathways.
• This involves using dimensionality reduction techniques (PCA, t-SNE, UMAP), clustering algorithms (Louvain, Leiden), differential gene expression analysis (such as MAST), advanced analysis, and pathway analysis.

Dimensionality Reduction

• Reduces the complexity of high-dimensional data to a lower dimension. Methods may include PCA (Principal Component Analysis), t-SNE (t-Distributed Stochastic Neighbor Embedding), and UMAP (Uniform Manifold Approximation and Projection) and are often used as a preliminary step.

Principal Component Analysis (PCA)

• PCA simplifies high-dimensional data by transforming it into a set of principal components.
• It is useful in scRNA-seq for noise reduction, preparation for clustering, and data visualization.
• It compresses the dataset, reduces complexity, and helps visualize relationships between cell populations after reduction in dimensionality.

t-SNE (t-Distributed Stochastic Neighbor Embedding)

• t-SNE is a non-linear dimensionality reduction technique, excellent for visualization of data.
• t-SNE focuses on local structure, emphasizing relationships among similar cells, with appropriate use in diverse single-cell datasets.
• It does not preserve global distances therefore is not ideal for all uses.

UMAP (Uniform Manifold Approximation and Projection)

• UMAP is another dimensionality reduction technique used for visualization of scRNA-seq data.
• UMAP preserves both local and global data structure offering a more holistic visualization.
• UMAP is more reproducible in results than t-SNE and is faster, making it suitable for big datasets.

Cluster Analysis in scRNA-Seq

• Clustering groups cells with similar gene expression profiles to identify distinct cell types or functional states, essential for understanding cellular diversity. Methods may include graph-based clustering algorithms like Louvain or hierarchical clustering.

Graph-Based Clustering

• Graph-based methods represent cells as nodes in a graph, connecting them based on gene expression similarity.
• Cells with similar expression are densely connected, forming clusters, like in the Louvain algorithm, useful in high-dimensional data.

Hierarchical Clustering

• A clustering technique that creates a dendrogram, a tree-like structure, representing cell relationships based on similarity.
• Agglomerative and divisive approaches progressively merge or split clusters based on similarity. Results reveal relationships at various levels of similarity.

Dimensionality Reduction VS Clustering

• Dimensionality reduction simplifies high-dimensional data for easier visualization and analysis before clustering begins.
• Clustering groups cells with similar gene expression profiles to identify cell types and their functional roles.

Differential Gene Expression (DGE)

• Identifies genes with different expression levels between cell clusters or conditions.
• Identifying marker genes helps distinguish cell types,
• Comparing conditions helps study gene expression changes between conditions or disease states.
• Pathway analysis links differentially expressed genes to specific pathways.

DEG Statistics

• Challenges in single-cell differential gene expression include the high variability and low counts in single cell data.
• Statistical methods, such as Wilcoxon Rank Sum Test, Likelihood Ratio Test (LRT), and MAST, are useful for addressing the challenges of analyzing differential gene expression. These methods accommodate zero values often present in scRNA-Seq analyses.

DEG Tools

• Seurat, DESeq2/edgeR,and MAST are useful tools for analyzing differential gene expression (DGE) in single-cell RNA sequencing (scRNA-seq) data. These tools handle various aspects of the data such as the size, technical variation, and integration into the workflow and provide suitable visualization tools.

Visualizing DEG

• Techniques for visualizing DEG include volcano plots (showing fold changes vs. significance), heatmaps (visualizing expression patterns), and dot plots (displaying expression levels), crucial for identifying marker genes and functional insights.

scRNA-Seq vs Bulk RNA-Seq

• scRNA-seq offers high cell-to-cell variability resolution but has more dropouts, lower read depth per cell, and may require sophisticated statistical and normalization methods.
• Bulk RNA-seq averages expression, reducing variability but may mask differences. scRNA-seq requires more complex and specialized tools and procedures for analysis and interpretation.

Pathway Analysis and Functional Enrichment

• Identifies biological pathways, cellular functions and processes associated with differentially expressed genes (DEGs) in scRNA-Seq datasets.
• Pathway analysis links DEGs to biochemical pathways (e.g., signaling, metabolic) which aids in interpreting the biological roles of specific cell states/types/conditions.
• Functional enrichment identifies overrepresented biological functions in DEG lists, using tools like DAVID, GSEA, KEGG and Reactome.

Pathway Enrichment

• Over-Representation Analysis (ORA) compares observed gene counts in pathways to what's expected by chance in scRNA-Seq analysis.
• Gene Set Enrichment Analysis (GSEA) ranks genes by their expression, useful for identifying pathways enriched at the top of the ranking and is sensitive to small changes in expression.
• Network-Based Analysis uses protein-protein interactions to identify functional modules within scRNA-Seq data.

Popular Enrichment Tools

• Tools like DAVID (Database for Annotation, Visualization, and Integrated Discovery), GSEA (Gene Set Enrichment Analysis), Reactome, and KEGG, along with ClusterProfiler offer enrichment analysis and pathway analysis for scRNA-seq data by helping analyze the functions and roles of different cell types in various contexts.

Interpreting Enrichment

• Focusing on biologically relevant pathways/genes that match known biology of cell types.
• Consider pathway redundancy.
• Analyze cell types and understand biological processes associated with them and generate hypotheses by understanding pathways and functions driving cellular behavior.

Cell Lineage Analysis

• Cell lineage analysis traces cell development and differentiation, identifying progression from stem cells/progenitors to fully differentiated cells using scRNA-seq.
• scRNA-Seq, using single-cell resolution, captures gene expression profiles at different stages, enabling researchers to identify and order cells along developmental or differentiation pathways.
• Techniques such as pseudotime analysis and lineage trees provide insights into developmental processes, identifying transitional cell states.

RNA Velocity

• RNA velocity predicts the "future state" of a cell based on the direction and rate of change in gene expression using unspliced and spliced mRNA.
• scRNA-seq snapshots are dynamic, unlike steady-state methods which use RNA velocities to study cellular transitions. RNA velocity provides a temporal perspective on cellular states and is useful in studying cellular development, differentiation, and disease progression.

Explore the fundamental concepts and advantages of single-cell RNA sequencing (scRNA-seq) compared to bulk RNA-seq. This quiz covers the definitions, techniques, and applications in cellular research and differentiation. Test your knowledge on data processing and sequencing platforms.