single cell transcriptomics

Question 1

heterogeneity in cell populations

Answer 1

• cell types
• somatic mutations
• cell cycle stage
• epigenetic modifications
• stochastic gene expression

Question 2

limitations of bulk assays

Answer 2

• assuming homogeneous relationships can lead you to the wrong conclusion
• rare cell types can become lost
• can’t see real time changes
  
  • need to order by differentiation progress, not time

Question 3

process of SCT

Answer 3

• isoolate cells
• lyse
• reverse transcribe and amplify cDNA
• qPCR or RNAseq
• up to 10,000s of genes in 10,000s of cells

Question 4

methods of single cell isolation

Answer 4

• low throughput:
  
  • manual/automated micropippetting
  • cytoplasmic aspiration
• high throughput:
  
  • FACS
  • microfluidics

Question 5

qPCR

Answer 5

• quantitative/real-time PCR
• gene specific PCR primers
• include housekeeping genes (GADPH)
• fluorescent dye to detect PCR product
• measure Ct value for each gene
  
  • threshold cycle number
• normalise data

Question 6

qPCR normalisation

Answer 6

• higher Ct means less cDNA
• arbitrary maximum Ct value
• calculate ΔCt for each gene
  
  • max - gene
  • higher Δ means more cDNA
• normalise with hk genes
  
  • assume hk expression constant
  • calculate gene ΔCt - hk ΔCt
• doubling cycles so subtraction not division

Question 7

RNAseq

Answer 7

• sequence cDNA library
• map reads to reference
• count read number for each gene
• need quality control
• can have coverage bias (5’/3’) in some protocols

Question 8

technical dropouts

Answer 8

• zero counts
• common
• when some mRNA not captured during reverse transcription
• capture efficiency:
  
  • % of mRNA molecules in cell lysate detected
  • often 10-20%
• more frequent in low expression genes
• varies between cells

Question 9

RNAseq normalisation

Answer 9

• convert raw read counts into expression levels per cell
• correct for cell to cell variation
  
  • in capture, amplification, sequencing efficiency
• method depends on protocol used
  
  • spike in or UMI

Question 10

extrinsic spike-ins

Answer 10

• add RNA of known sequence and quantity to lysate
• internal control
• equal quantity in each lysate
• normalise counts by number of reads mapped by spike in RNA
• assumes same capture, amplification and sequencing efficiencies
• be cautious:
  
  • no 5’ cap or polyA tail

Question 11

UMIs

Answer 11

• unique molecular identifiers
• barcode on each cDNA moelcule
• 6-10 nt added before amplification
• track how much of amplified DNA comes form original molecule
• count number of unique UMIs associated with each gene
• assume library sequenced to saturation
• corrects for variation in amplification efficiency but not other sources
  
  • e.g. reverse transcription

Question 12

normalisation without spike in or UMI

Answer 12

• same as used by bulk RNAseq data
• assume hk gene expression or total mRNA content the same
• normalise read counts by hk expression/total mRNA
• cna also combine techniques

Question 13

SC data analysis techniques

Answer 13

• clustering
• dimensionality reduction
• differential expression
• pseudotemporal ordering
• network interference

Question 14

single cell clustering

Answer 14

• cluster by trancriptomic profile to:
  
  • analyse sub-population structure
  • identify cell sub-types/rare cell types
• cluster by cell expression states to:
  
  • identify co-varying genes

Question 15

SC clustering methods

Answer 15

• partitional
  
  • produces disjoint groups
  • k-means
• hierarchical clustering
  
  • divisive or agglomerative
  • hierarchical tree
  • can provide more information

Question 16

k-means clustering

Answer 16

• algorithms putting data points into k clusters
• cluster points with the most similar mean average
• have to choose k

Question 17

bi-clustering

Answer 17

• allows clustering by genes and cells simultaneously
• find genes that behave similarly within a cell cluster
• can give better resolution
• only some subsets are informative

Question 18

dimensionality reduction

Answer 18

• transform to lower dimensional space e.g. 3D to 2D
• easier to visualise data and detect patterns
• often before clustering
  
  • distance can behave non-intuitively in high dimensions
• many algorithms with different assumptions
  
  • PCA

Question 19

PCA

Answer 19

• linear transformation of uncorrelated principal components
• PCs:
  
  • orthogonal
  • ordered by contribution to variance in data
  • weighted sum or original dimensions
• PC1:
  
  • vector through dataset giving largest amount of variation
  • PC2 gives second largest
• shows which genes contribute most to heterogeneity
• can combine with clustering to analyse distinct populations

Question 20

differential expression

Answer 20

• aim to detect difference in gene expression levels or distribution between 2 cell populations
• statistical tests
  
  • T-test, Mann-Whitney
  • specialised methods needed for noise/dropouts
    
    • more noise in SC than bulk
• multiple testing corrections

Question 21

GO enrichment

Answer 21

• gene ontology
  
  • describes gene functions and relationships between them
  • molecular funciton, cellular component, biological process
• identify terms over/underrepresented in a given set of genes
• select input list of genes of interest
  
  • DE genes, bi-clustering genes

Question 22

GO output

Answer 22

• calculate probability of seeing observed sample frequency by chance given the background frequency for each term
  
  • sample frequency = no of genes annotated to that term in the input
  • background frequency = no of genes annotated to a term in the background set
  • background set = all genes in the genome
• identify which terms appear more frequently than expected

Question 23

pseudotemporal ordering

Answer 23

• aim to infer gene expression dynamics from snapshot data
  
  • true temporal data unavailable
  • measurement destroys cells by lysis
• cells ordered by progress through a biological process
  
  • differentiation, response to stimuli
• sampling time may not correlate well with stages
  
  • asynchrony
• assumes cells follow same response or differentiation path
• monocle algorithm

Question 24

gene regulatory network inference

Answer 24

• aim to ocnstruct a network graph where nodes represent genes and edges indicate regulatory interactions
• assumes that a strong statistical relationship between 2 gene expression profiles indicates a potential functional relationship
  
  • correlation vs mutual information
• linked groups of genes indicate that they undergo coordinated changes in expression
• some correlations may be confounding factors e.g. cell cycle
• may need to choose subpopulations

Question 25

correlation vs mutual information

Answer 25

• correlation
  
  • most common measure for strength of a statistical relationship
• mutual information
  
  • alternative that can identify non-linear relationships