Describe single-cell sequencing technologies

Overview

Single-cell sequencing is a revolutionary technique that measures gene expression (or other molecular features) at the resolution of individual cells, rather than averaging across millions of cells in bulk tissue. This allows us to discover rare cell types, track developmental trajectories, and understand cellular heterogeneity that bulk methods completely miss.

[!intuition] The Core Problem & Insight

Why do we need single-cell resolution?

Imagine you're trying to understand a symphony orchestra by recording all instruments at once into a single microphone. You'd hear the average sound, but you'd miss:

• The rare triangle player (rare cell types)
• How the violins differ from celos (cell-type heterogeneity)
• That one violin is slightly out of tune (individual cell variation)

Bulk RNA sequencing is like that single microphone—it averages expression across millions of cells. If99% of cells are type A and 1% are rare stem cells, the stem cell signal drowns in the noise.

Single-cell sequencing isolates individual cells, tags each cell's RNA with a unique barcode, then sequences them separately. Now you can:

1. Identify rare populations (e.g., circulating tumor cells at 1 in 10,000)
2. Resolve cell states along a continuum (e.g., stem → progenitor → differentiated)
3. Map tissue architecture at cellular resolution

[!definition] Key Concepts

Single-Cell RNA Sequencing (scRNA-seq)

A method that profiles the transcriptome (all mRNA molecules) of individual cells. For each cell, you obtain a list of which genes are expressed and at what levels.

Output: A gene × cell matrix where entry (i,j) = number of mRNA copies of gene i in cell j.

Cell Barcoding

Each cell'sDNA is tagged with a unique molecular barcode (typically 10-16 nucleotides) so that millions of cells can be pooled and sequenced together, then computationally separated afterward.

Unique Molecular Identifiers (UMIs)

Random short sequences (4-10 nt) added to each individual mRNA molecule before amplification. This allows you to count the original number of mRNA molecules (not amplified copies), correcting for PCR bias.

Why? PCR amplifies some molecules more than others. Without UMIs, 1 original mRNA that amplified 100× looks like 100 original mRNAs.

[!formula] The Measurement Pipeline: From Cell to Count Matrix

Step 1: Cell Isolation & Capture

Goal: Get individual cells into separate reaction compartments.

Methods:

1. Fluorescence-Activated Cell Sorting (FACS):

   • Sort cells one-by-one into 96-well or 384-well plates
   • High purity, but low throughput (~1000 cells/experiment)

2. Microfluidic Droplet-Based Systems (e.g., 10x Genomics, Drop-seq):

   • Encapsulate each cell in a nanoliter droplet with a bead carrying unique barcodes
   • Throughput: 10,000-100,000 cells per run
   • How it works:
     • Cells flow through one channel, barcoded beads through another
     • At junction, oil creates droplets: ideally 1 cell + 1 bead per droplet (following Poisson statistics, most droplets are empty, some have 1 cell, rare doublets have 2)

Poisson Statistics: If you aim for an average of λ = 0.1 cells per droplet: $P(\text{exactly 1 cell}) = \lambda e^{-\lambda} \approx 0.09$ $P(\text{doublet, 2 cells}) = \frac{\lambda^2}{2} e^{-\lambda} \approx 0.005$

Why this loading rate? Balance: higher λ → more cells captured, but more doublets (which confound analysis).

Step 2: Cell Lysis & mRNA Capture

Inside each droplet:

1. Cell lyses (bursts open)
2. mRNA molecules hybridize to oligo-dT primers on the barcoded bead
3. The bead oligo structure: $\text{5'─ [Illumina adapter] ─ [Cell Barcode] ─ [UMI] ─ [oligo-dT] ─3'}$

Why oligo-dT? mRNA has a poly-A tail; oligo-dT (string of T nucleotides) binds specifically to poly-A, capturing mRNA (not rRNA or tRNA).

Step 3: Reverse Transcription

In the droplet, reverse transcriptase synthesizes cDNA from mRNA using the bead oligo as primer:

$\text{mRNA-AA...} + \text{TT-[UMI]-[Barcode]-[Adapter]} \xrightarrow{\text{RT}} \text{cDNA tagged with barcode+UMI}$

Result: Each cDNA molecule now carries:

• Cell barcode: which cell it came from
• UMI: which original mRNA molecule it came from
• Adapter: for sequencing

Step 4: Pooling, Amplification & Sequencing

1. Break droplets, pool all cDNA together (millions of barcoded molecules in one tube)
2. PCR amplify the cDNA library
3. Sequence using Illumina short-read sequencing:
   • Read 1: Cell barcode + UMI (26 bp total)
   • Read 2: cDNA sequence (maps to a gene)

Step 5: Computational Deconvolution

From raw reads to count matrix:

1. Demultiplex: Group reads by cell barcode → assign reads to cells
2. Align: Map Read 2 to the reference genome/transcriptome → identify which gene
3. Collapse UMIs: For each (cell, gene) pair, count unique UMIs (not total reads)
   • If gene ACTB in cell #1572 has 50 reads but only 10 unique UMIs → count = 10original mRNAs

Output: Gene expression matrix:

where $G$ = number of genes (~20,000 human), $C$ = number of cells (10,000-100,000).

Why this matters: $X_{ij}$ = UMI count for gene $i$ in cell $j$. This is the dataset you analyze to find cell types, trajectories, etc.

[!example] Worked Example 1: Calculating Doublet Rate

Scenario: You're running a 10x Genomics experiment targeting 10,000 cells. The system loads cells into droplets with mean λ = 0.1 cells/droplet.

Question: What fraction of droplets with≥1 cell are doublets?

Solution:

The number of cells per droplet follows a Poisson distribution: $P(k \text{ cells}) = \frac{\lambda^k e^{-\lambda}}{k!}$

For λ = 0.1: $P(0) = e^{-0.1} = 0.9048$ $P(1) = 0.1 \cdot e^{-0.1} = 0.0905$ $P(2) = \frac{0.1^2}{2} e^{-0.1} = 0.0045$

Doublet rate among captured cells: $\text{Doublet rate} = \frac{P(\text{doublet})}{P(\geq 1 \text{ cell})} = \frac{P(2)}{P(1) + P(2) P(3) + ...}$

Since $P(\geq 1) = 1 - P(0) = 0.0952$: $\text{Doublet rate} \approx \frac{0.0045}{0.0952} = 4.7\%$

Why this step? You need to know what fraction of your "cells" are actually two cells stuck together, which creates false cell types in downstream analysis. At λ = 0.1, ~5% doublets is acceptable; at λ = 0.3, doublets jump to ~15%, which is problematic.

[!example] Worked Example 2: UMI Correction

Scenario: For gene CD8A in cell #412, sequencing produced:

• UMI ACGT: 50 reads
• UMI ACGG: 30 reads (Haming distance 1 from ACGT)
• UMI TTCA: 25 reads

Question: How many original CD8A mRNAs were in this cell?

Solution:

Naive count (wrong): 50 + 30 + 25 = 105 reads → 3 UMIs → report 3 mRNAs.

Problem: UMI ACGG is 1 nucleotide away from ACGT. This is likely a sequencing error, not a real different molecule.

UMI correction algorithm (used by Cell Ranger):

1. Rank UMIs by read count: ACGT (50), ACGG (30), TTCA (25)
2. For each UMI, check if a more-abundant UMI exists within Hamming distance ≤1
3. ACGG → collapse into ACGT
4. TTCA → no close high-count UMI, keep separate

Corrected count: 2 unique UMIs → 2 original mRNAs.

Why this matters: Without UMI correction, sequencing errors inflate counts by ~30%. With correction, you get true molecular counts.

[!example] Worked Example 3: Discovering Rare Cell Types

Biological Question: A mouse brain has ~1 million cells. You suspect a rare neuron subtype exists at 0.1% frequency (1000 cells). Can scRNA-seq find it?

Bulk RNA-seq: Averages across 1 million cells. The rare subtype's unique marker gene RARE1 expressed at 1000 TPM in rare cells,0 TPM elsewhere.

Bulk average: $\text{Bulk signal} = 0.001 \times 1000 + 0.999 \times 0 = 1 \text{ TPM}$

Problem: 1 TPM is below noise floor; the rare cell type is invisible.

scRNA-seq: Sequence 50,000 cells (5% of tissue).

Expected rare cells captured: $5000 \times 0.001 = 50 \text{ cells}$

In the gene expression matrix, these50 cells show:

• RARE1 expression >> 0 (while 49,950 other cells have RARE1 = 0)
• Clustering algorithms (e.g., Louvain) group these 50 cells into a distinct cluster

Result: You discover the rare subtype, can profile its unique transcriptome, and validate with marker genes.

Why scRNA-seq wins: Rare signals aren't diluted; each cell "votes" independently.

[!mistake] Common Mistakes & Misconceptions

Mistake 1: "More sequencing depth per cell is always better"

Why it feels right: More reads → more genes detected → better data.

The fix: There's a saturation point. Human cells express ~10,000-15,000 unique mRNAs. Once you've sequenced ~50,000 reads/cell, you've captured most abundant and mid-abundance transcripts. Going to 200,000 reads/cell:

• Increases cost4×
• Detects only ultra-rare transcripts (which may be noise)
• Doesn't help you discover new cell types

Better strategy: Sequence 10,000 reads/cell but sample more cells (10,000 → 50,000 cells). You'll capture more biological diversity.

Rule of thumb:

• Shallow sequencing (5-10k reads/cell): cell-type discovery
• Deep sequencing (50-100k reads/cell): rare isoform detection, velocity analysis

Mistake 2: "Doublets don't matter if they're only 5%"

Why it feels right: 5% is small; 95% of data is clean.

The fix: Doublets create false cell types. If cell type A (marker: GeneA) and cell type B (marker: GeneB) form a doublet, thelet expresses both GeneA and GeneB.

Clustering algorithms see this as:

• A novel "hybrid" cell type (wrong!)
• Or split one true cell type into two (worse!)

Real impact: In a study that found "20 cell types," 3 turned out to be doublet artifacts. Always run doublet-detection software (Scrublet, DoubletFinder).

Mistake 3: "UMI count = absolute mRNA count"

Why it feels right: UMIs tag original molecules, so UMI count should equal molecule count.

The fix: UMI counts are relative, not absolute. Sources of bias:

1. Capture efficiency: Only ~10-20% of mRNAs in a cell get captured and reverse-transcribed
2. mRNA length bias: Longer mRNAs are easier to capture (more opportunities for oligo-dT binding)
3. Cell size: Larger cells have more total mRNA, so same gene expression level → higher UMI count

Correct interpretation: Compare UMI counts within a gene across cells (relative expression), but don't compare counts between genes as if they're molecules/cell.

To get absolute counts, you'd need spike-in standards (which most protocols skip for cost).

[!recall]- Explain to a 12-Year-Old

Imagine you want to know what every person in a huge stadium is saying, but you can only hear the average of all voices together—total noise! That's what scientists faced with bulk RNA sequencing: they'd mash up millions of cells and measure the average genes, missing all the interesting rare cells and differences.

Single-cell sequencing is like giving each person in the stadium a unique name tag (the barcode), then recording everyone separately. Now you can:

• Find the one person yelling something different (rare cell type)
• See that people in section A are chering while section B is boing (different cell states)
• Track how someone's mood changes from the start to end of the game (cell trajectory)

The trick is: You use tiny droplets (like soap bubbles) to trap one cell per bubble, along with a bead that stamps that cell's RNA with its unique ID. Then you sequence everything together and use the IDs to figure out which RNA came from which cell. It's like sorting a million mixed-up LEGO bricks back into their original sets by reading the set number printed on each brick!

[!mnemonic] DROPLET Mnemonic

Droplets isolate cells
Reverse transcription tags RNA
Oligo-dT captures mRNA
Poisson statistics control doublets
Lysis releases molecules
Each cell gets unique barcode
Thousands of cells, one experiment

Connections

• Bulk RNA Sequencing: Compare with single-cell resolution vs. population averages
• Clustering Algorithms: Louvain, Leiden for identifying cell types from scRNA-seq
• Dimensionality Reduction: t-SNE, UMAP for visualizing high-dimensional single-cell data
• Trajectory Inference: Pseudotime analysis to order cells along developmental paths
• Cell Type Annotation: Using marker genes and databases to label clusters
• Doublet Detection: Computational methods to identify and remove doublets
• Gene Expression Quantification: TPM, FPKM vs. UMI counts
• Next-Generation Sequencing: Illumina platform underlying scRNA-seq
• Poisson Distribution: Models cell loading in droplet-based systems
• Spatial Transcriptomics: Adds spatial coordinates to single-cell expression

#flashcards/biology

What is single-cell RNA sequencing (scRNA-seq)? :: A technique that measures gene expression at individual cell resolution by isolating cells, tagging each cell's RNA with unique barcodes, and sequencing them separately to create a gene × cell expression matrix.

Why can't bulk RNA-seq detect rare cell types at 0.1% frequency? :: Bulk sequencing averages expression across all cells, so rare cell signals (0.1% of cells) are diluted 1000-fold and fall below the detection noise floor, making them invisible.

Concept Map

Hinglish (regional understanding)

Hinglish (regional understanding)