Describe single-cell sequencing technologies

6.5.10 · Biology › Systems Biology & Frontiers

Overview

Single-cell sequencing ek revolutionary technique hai jo gene expression (ya doosre molecular features) ko individual cells ke level par measure karti hai, na ki bulk tissue mein millions of cells ka average leke. Isse hum rare cell types discover kar sakte hain, developmental trajectories track kar sakte hain, aur cellular heterogeneity samajh sakte hain jo bulk methods bilkul miss kar dete hain.

[!intuition] Core Problem aur Insight

Hume single-cell resolution ki zaroorat kyun hai?

Imagine karo tum ek symphony orchestra ko samajhna chahte ho aur ek single microphone mein saare instruments ek saath record kar rahe ho. Tum average sound sunoge, lekin miss karoge:

• Rare triangle player ko (rare cell types)
• Yeh ki violins, cellos se kaise alag hain (cell-type heterogeneity)
• Woh ek violin jo thodi si out of tune hai (individual cell variation)

Bulk RNA sequencing us single microphone jaisi hai — woh millions of cells mein expression ko average kar deti hai. Agar 99% cells type A hain aur 1% rare stem cells hain, toh stem cell signal noise mein dub jaata hai.

Single-cell sequencing individual cells ko isolate karti hai, har cell ki RNA ko ek unique barcode se tag karti hai, phir unhe separately sequence karti hai. Ab tum kar sakte ho:

1. Rare populations identify karna (jaise, circulating tumor cells 1 in 10,000 par)
2. Cell states resolve karna ek continuum ke saath (jaise, stem → progenitor → differentiated)
3. Tissue architecture map karna cellular resolution par

[!definition] Key Concepts

Single-Cell RNA Sequencing (scRNA-seq)

Ek method jo individual cells ke transcriptome (saare mRNA molecules) ko profile karta hai. Har cell ke liye, tum dekhte ho ki kaun se genes express ho rahe hain aur kis level par.

Output: Ek gene × cell matrix jisme entry (i,j) = cell j mein gene i ke mRNA copies ki sankhya.

Cell Barcoding

Har cell ki DNA ko ek unique molecular barcode se tag kiya jaata hai (typically 10-16 nucleotides) taaki millions of cells ko pool karke ek saath sequence kiya ja sake, phir computationally alag kiya ja sake.

Unique Molecular Identifiers (UMIs)

Random short sequences (4-10 nt) jo har individual mRNA molecule mein amplification se pehle add kiye jaate hain. Isse tum original mRNA molecules ki sankhya count kar sakte ho (amplified copies nahi), jo PCR bias ko correct karta hai.

Kyun? PCR kuch molecules ko doosron se zyada amplify karta hai. UMIs ke bina, 1 original mRNA jo 100× amplify hua woh 100 original mRNAs jaisa lagta hai.

[!formula] Measurement Pipeline: Cell se Count Matrix Tak

Step 1: Cell Isolation aur Capture

Goal: Individual cells ko alag reaction compartments mein lane ka.

Methods:

1. Fluorescence-Activated Cell Sorting (FACS):

   • Cells ko ek-ek karke 96-well ya 384-well plates mein sort karo
   • High purity, lekin low throughput (~1000 cells/experiment)

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

   • Har cell ko ek nanoliter droplet mein unique barcodes wale bead ke saath encapsulate karo
   • Throughput: 10,000-100,000 cells per run
   • Yeh kaise kaam karta hai:
     • Cells ek channel se flow karte hain, barcoded beads doosre se
     • Junction par, oil droplets banata hai: ideally 1 cell + 1 bead per droplet (Poisson statistics follow karte hue, zyaatar droplets empty hote hain, kuch mein 1 cell hoti hai, rare doublets mein 2 hoti hain)

Poisson Statistics: Agar tum average λ = 0.1 cells per droplet target karo: $P(\text{exactly 1 cell})=\lambda e^{-\lambda }\approx 0.09$ $P(\text{doublet, 2 cells})=\frac{{\lambda }^2}{2}{e}^{-\lambda }\approx 0.005$

Yeh loading rate kyun? Balance: zyada λ → zyada cells capture, lekin zyada doublets (jo analysis ko confound karte hain).

Step 2: Cell Lysis aur mRNA Capture

Har droplet ke andar:

1. Cell lyse hoti hai (burst ho jaati hai)
2. mRNA molecules, barcoded bead par oligo-dT primers se hybridize hote hain
3. Bead oligo ka structure: $\text{5’─ [Illumina adapter] ─ [Cell Barcode] ─ [UMI] ─ [oligo-dT] ─3’}$

Oligo-dT kyun? mRNA mein poly-A tail hoti hai; oligo-dT (T nucleotides ki string) specifically poly-A se bind karta hai, mRNA capture karta hai (rRNA ya tRNA nahi).

Step 3: Reverse Transcription

Droplet mein, reverse transcriptase, bead oligo ko primer ke roop mein use karke mRNA se cDNA synthesize karta hai:

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

Result: Har cDNA molecule ab carry karta hai:

• Cell barcode: yeh kis cell se aaya
• UMI: yeh kis original mRNA molecule se aaya
• Adapter: sequencing ke liye

Step 4: Pooling, Amplification aur Sequencing

1. Droplets tod do, saara cDNA ek saath pool karo (ek tube mein millions of barcoded molecules)
2. PCR amplify karo cDNA library ko
3. Sequence karo Illumina short-read sequencing se:
   • Read 1: Cell barcode + UMI (26 bp total)
   • Read 2: cDNA sequence (ek gene pe map hota hai)

Step 5: Computational Deconvolution

Raw reads se count matrix tak:

1. Demultiplex: Reads ko cell barcode se group karo → reads ko cells ko assign karo
2. Align: Read 2 ko reference genome/transcriptome se map karo → identify karo kaun sa gene hai
3. UMIs collapse karo: Har (cell, gene) pair ke liye, unique UMIs count karo (total reads nahi)
   • Agar cell #1572 mein gene ACTB ke 50 reads hain lekin sirf 10 unique UMIs → count = 10 original mRNAs

Output: Gene expression matrix:

jahaan $G$ = genes ki sankhya (~20,000 human), $C$ = cells ki sankhya (10,000-100,000).

Yeh kyun matter karta hai: $X_{ij}$ = gene $i$ ka UMI count cell $j$ mein. Yahi woh dataset hai jise tum cell types, trajectories, etc. dhundne ke liye analyze karte ho.

[!example] Worked Example 1: Doublet Rate Calculate Karna

Scenario: Tum 10x Genomics experiment run kar rahe ho jisme 10,000 cells target kar rahe ho. System cells ko droplets mein load karta hai mean λ = 0.1 cells/droplet ke saath.

Question: ≥1 cell wale droplets mein se kitne fraction doublets hain?

Solution:

Har droplet mein cells ki sankhya Poisson distribution follow karti hai: $P(k\text{ cells})=\frac{{\lambda }^k{e}^{-\lambda }}{k!}$

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

Captured cells mein doublet rate: $\text{Doublet rate}=\frac{P(\text{doublet})}{P(\ge 1\text{ cell})}=\frac{P(2)}{P(1)+P(2)P(3)+...}$

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

Yeh step kyun? Tume jaanna hai ki tumhare kitne "cells" actually do cells hain jo ek saath chipke hain, jo downstream analysis mein false cell types banate hain. λ = 0.1 par, ~5% doublets acceptable hai; λ = 0.3 par, doublets ~15% tak jump kar jaate hain, jo problematic hai.

[!example] Worked Example 2: UMI Correction

Scenario: Cell #412 mein gene CD8A ke liye, sequencing ne produce kiya:

• UMI ACGT: 50 reads
• UMI ACGG: 30 reads (ACGT se Hamming distance 1)
• UMI TTCA: 25 reads

Question: Is cell mein kitne original CD8A mRNAs the?

Solution:

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

Problem: UMI ACGG, ACGT se 1 nucleotide door hai. Yeh likely ek sequencing error hai, real alag molecule nahi.

UMI correction algorithm (Cell Ranger use karta hai):

1. UMIs ko read count se rank karo: ACGT (50), ACGG (30), TTCA (25)
2. Har UMI ke liye, check karo ki Hamming distance ≤1 mein koi zyada abundant UMI exist karta hai
3. ACGG → ACGT mein collapse karo
4. TTCA → koi close high-count UMI nahi, alag rakho

Corrected count: 2 unique UMIs → 2 original mRNAs.

Yeh kyun matter karta hai: UMI correction ke bina, sequencing errors counts ko ~30% inflate kar deti hain. Correction ke saath, tume true molecular counts milte hain.

[!example] Worked Example 3: Rare Cell Types Discover Karna

Biological Question: Ek mouse brain mein ~1 million cells hain. Tumhe lagta hai ki ek rare neuron subtype 0.1% frequency par exist karta hai (1000 cells). Kya scRNA-seq use dhundh sakta hai?

Bulk RNA-seq: 1 million cells mein average karta hai. Rare subtype ka unique marker gene RARE1 rare cells mein 1000 TPM par express hota hai, baaki jagah 0 TPM.

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

Problem: 1 TPM noise floor se neeche hai; rare cell type invisible hai.

scRNA-seq: 50,000 cells sequence karo (tissue ka 5%).

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

Gene expression matrix mein, yeh 50 cells dikhate hain:

• RARE1 expression >> 0 (jabki 49,950 doosri cells mein RARE1 = 0)
• Clustering algorithms (jaise, Louvain) in 50 cells ko ek distinct cluster mein group karte hain

Result: Tum rare subtype discover karte ho, uska unique transcriptome profile kar sakte ho, aur marker genes se validate kar sakte ho.

scRNA-seq kyun jeetta hai: Rare signals dilute nahi hote; har cell "vote" karta hai independently.

[!mistake] Common Mistakes aur Misconceptions

Mistake 1: "Har cell ke liye zyada sequencing depth hamesha better hai"

Yeh kyun sahi lagta hai: Zyada reads → zyada genes detected → better data.

Fix: Ek saturation point hota hai. Human cells ~10,000-15,000 unique mRNAs express karte hain. Ek baar ~50,000 reads/cell sequence karne ke baad, tumne zyaatar abundant aur mid-abundance transcripts capture kar liye hain. 200,000 reads/cell tak jaana:

• Cost 4× badhata hai
• Sirf ultra-rare transcripts detect karta hai (jo noise ho sakti hai)
• Naye cell types discover karne mein help nahi karta

Better strategy: 10,000 reads/cell sequence karo lekin zyada cells sample karo (10,000 → 50,000 cells). Tum zyada biological diversity capture karoge.

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 matter nahi karte agar woh sirf 5% hain"

Yeh kyun sahi lagta hai: 5% chhota hai; 95% data clean hai.

Fix: Doublets false cell types banate hain. Agar cell type A (marker: GeneA) aur cell type B (marker: GeneB) ek doublet banaate hain, toh woh doublet dono GeneA aur GeneB express karta hai.

Clustering algorithms ise dekhte hain:

• Ek novel "hybrid" cell type (galat!)
• Ya ek true cell type ko do mein split kar dete hain (aur bura!)

Real impact: Ek study mein jo "20 cell types" mili, 3 doublet artifacts nikle. Hamesha doublet-detection software chalao (Scrublet, DoubletFinder).

Mistake 3: "UMI count = absolute mRNA count"

Yeh kyun sahi lagta hai: UMIs original molecules ko tag karte hain, isliye UMI count molecule count ke barabar hona chahiye.

Fix: UMI counts relative hain, absolute nahi. Bias ke sources:

1. Capture efficiency: Ek cell mein sirf ~10-20% mRNAs hi capture aur reverse-transcribe hote hain
2. mRNA length bias: Lambe mRNAs zyada aasani se capture hote hain (oligo-dT binding ke zyada opportunities)
3. Cell size: Badi cells mein zyada total mRNA hota hai, isliye same gene expression level → zyada UMI count

Sahi interpretation: UMI counts ko cells mein ek gene ke andar compare karo (relative expression), lekin genes ke beech counts ko molecules/cell ki tarah compare mat karo.

Absolute counts ke liye, tumhe spike-in standards chahiye honge (jise zyaatar protocols cost ki wajah se skip karte hain).

[!recall]- Ek 12-Saal-Ke Bachche Ko Samjhao

Imagine karo tum ek bade stadium mein har ek insaan kya bol raha hai jaanna chahte ho, lekin tum sirf saari awazon ka average ek saath sun sakte ho — total shor! Yahi scientists ka problem tha bulk RNA sequencing ke saath: woh millions of cells ko mash up karte aur average genes measure karte, saare interesting rare cells aur differences miss karte.

Single-cell sequencing aisa hai jaise stadium mein har ek insaan ko ek unique name tag dena (woh barcode), phir har ek ko alag record karna. Ab tum kar sakte ho:

• Woh ek insaan dhundna jo kuch alag chilla raha hai (rare cell type)
• Yeh dekhna ki section A ke log cheer kar rahe hain jabki section B boo kar raha hai (different cell states)
• Track karna ki kisi ki mood game ke shuru se end tak kaise badli (cell trajectory)

Trick hai: Tum tiny droplets (saabun ke bubbles jaisi) use karte ho ek cell per bubble mein band karne ke liye, ek bead ke saath jo us cell ki RNA par uska unique ID stamp karta hai. Phir tum sab kuch ek saath sequence karte ho aur IDs use karte ho yeh figure out karne ke liye ki kaun sa RNA kaun si cell se aaya. Yeh aisa hai jaise ek million mixed-up LEGO bricks ko wapas unke original sets mein sort karna har brick par printed set number padh ke!

[!mnemonic] DROPLET Mnemonic

Droplets cells ko isolate karte hain Reverse transcription RNA ko tag karta hai Oligo-dT mRNA capture karta hai Poisson statistics doublets control karte hain Lysis molecules release karta hai Each cell ko unique barcode milta hai Thousands of cells, ek experiment

Connections

• Bulk RNA Sequencing: Single-cell resolution vs. population averages se compare karo
• Clustering Algorithms: scRNA-seq se cell types identify karne ke liye Louvain, Leiden
• Dimensionality Reduction: High-dimensional single-cell data visualize karne ke liye t-SNE, UMAP
• Trajectory Inference: Cells ko developmental paths ke saath order karne ke liye Pseudotime analysis
• Cell Type Annotation: Clusters label karne ke liye marker genes aur databases use karna
• Doublet Detection: Doublets identify aur remove karne ke liye computational methods
• Gene Expression Quantification: TPM, FPKM vs. UMI counts
• Next-Generation Sequencing: scRNA-seq ke underlying Illumina platform
• Poisson Distribution: Droplet-based systems mein cell loading ko model karta hai
• Spatial Transcriptomics: Single-cell expression mein spatial coordinates add karta hai

#flashcards/biology

Single-cell RNA sequencing (scRNA-seq) kya hai? :: Ek technique jo gene expression ko individual cell resolution par measure karti hai — cells ko isolate karke, har cell ki RNA ko unique barcodes se tag karke, aur unhe separately sequence karke ek gene × cell expression matrix banati hai.

Bulk RNA-seq 0.1% frequency par rare cell types kyun detect nahi kar sakta? :: Bulk sequencing saari cells mein expression ko average kar deta hai, isliye rare cell signals (0.1% cells) 1000-fold dilute ho jaate hain aur detection noise floor se neeche chale jaate hain, inhe invisible bana dete hain.

Concept Map