List

Advantages

+ characterization of cellular heterogeneity within a population. (identify rare cell types, distinguish different cell states, capture cell-to-cell variability in gene expression)

+ discovery of novel cell types or subpopulations

+ can capture gene expression changes over time

Disadvantages

- Higher cost and lower throughput

- more complex data analysis

- Increased technical noise

• sequencing depth -> Between sample comparison

• gene length -> Within sample comparison

TPM

• divide by length of gene, constant 10^6, total number of reads

• Use: within or between sample groups

CPM (only for depth)

• divide by total number of reads, constant 10^6

• Use: between sample groups

RPKM

• divide by total number of reads, constant 10^6, length of a gene

• within sample groups

FPKM

divide alle RPKM by 2

normalize each sample independently

• RNA Composition

• Batch effects

1. Cell Barcodes

• label and differentiate RNA molecules originating from different cells within a mixed population.

• distinguish and assign RNA reads to specific cells, enabling cell-level analysis in scRNA-seq.

2. Unique Molecular Identifier (UMI)

• used to distinguish individual RNA molecules within a cell.

• utilized to identify and quantify individual RNA molecules within a cell, improving the accuracy of gene expression quantification.

Summary

Both barcode types are essential components of scRNA-seq library preparation and play critical roles in demultiplexing and data analysis.

The droplet method, also known as droplet-based scRNAseq is a technique used in single-cell genomics to analyze the gene expression profiles of individual cells.

Steps

1. Cell suspension & barcoding

2. Droplet Generation

3. Emulsion

4. Reaction in Droplets

5. Reaction after Demulsification

6. Down Stream Applications etc.

1. Preprocessing

   • control, normalization, and feature selection

2. Visualization

   • PCA, t-SNE, UMAP

3. Analysis

   • Cluster Analysis, Trajectory Analysis

• cell cluster -> first result of any single-cell-anlysis

• => infer the identity of member cells

• Clusters groups cells based on the similarity of their gene expression profiles

• Expression profile similarity - determined by distance metrics (takes dimensionality reduced representations as input)

• Two approaches to generate cell clusters from similarity scores:

  • clustering algorithms

  • community detection methods

Cells with…

• unexpected high count

• Large #detected genes

help prevent PCR-bias

Cause of Bias?

• sequencing depth

• library efficiency

• amplification bias

How to correct?

TBD

Model expression mean vs. variance over all genes -> genes w/ similar mean are estimated together for their variance

empirical Bayes estimation:

Step 1: creates a pseudo-reference sample (row-wise geometric mean)

Step 2: calculates ratio of each sample to the reference

Step 3: calculate the normalization factor for each sample (size factor)

Negative binomial distribution

-> overdispersion

Read counts descriped by Poisson distribution (mean and variance are equal)

-> Not the case for RNAseq-counts (overdispersion)

=> negativ binomial distribution - takes this into account through a dispersion

parameter alpha

transcript is present at slightly different levels in each sample

• STAR is a alignment tool for RNAseq data.

• Usage of uncompressed suffix array to search for MMPs

1. maximal mappable prefix (MMP) is determined for each aligned read

2. Seed: search for unmapped suffix that can be in another exon

3. Extend: no exact match for seed found -> extend previous seed; if no good alignment -> remove suffix from read

4. stich seeds togehter based on (a) proximity to a set of anchors or (b) best on best alignment of read

Goal: Estimate transcript abundances

• ==> Classify diseases, understand expression changes, track cancer progression

• Estimation but accurate

• faster and more efficient than alignment based approaches

• Tools: Kallisto and Salmon

1. Identify likely miRNA-gene pairs

2. Indetify ceRNA pairs (<-> shared miRNAs) and calculate sensitivity correlation

3. Use SPONGE null model to infere p-values of the significance of the interaction

• Gene-gene correlation

• sample size

• several miRNAs regulate many transcripts

IN

• Reference transcriptome

• RNA-Seq reads from experiment

Steps

1. Indexing / Hashing of k-mers

   • Construction of hash table of k-mers to contigs and their position within

   • Skipping of redundant k-mers in same k-compatibility class

   • Intersection of constituent k-mers => k-compatibility class of read

2. Pseudoalignment

   • lookup of k-compatibility class for each k-mer in kallisto index, intersecting the k-compatibility classes

   • K-mer hashing is strand agnostic

   • Optimization: All k-mers in a contig of the de Bruijn graph have the same k-compatibility class

   • => For each k-mer lookup, find distance to junctions at the end of contig ==> skip k-mers up to that distance

3. Quatification

OUT

Kallisto-Index, Quantification of RNA-Seq samples (transcript Pseudoalignment abundances)

Large non-coding RNAs

• lncRNA

• eRNA

• circRNA

Small non-coding RNAs

• miRNA

• siRNA

• miRNAs are 19-22 nucleotide long molecules

• key regulators of gene expression -> each miRNA can potentially regulate hundreds of genes

Competing endogenous RNA hypothesis == SPONGE hypothesis

• RNA can compete for binding to miRNAs through shared MREs (miRNA response elements) present in their sequences

  => influence availability of miRNAs and affect the expression of target messenger RNAs (mRNAs)

• suggests that when non-coding RNAs and coding RNAs contain similar MREs, they can act as "sponges" for miRNAs

• this interaction among RNA molecules => complex regulatory network

  => abundance of one RNA => influence the expression of other RNAs (by competing for shared miRNAs)

ceRNA <=> Key for hidden RNA language and found everywhere

• Microarray

• Nanopore & -drop

• Illumina HiSeq

• transcript discovery

• gene quantification

• expression profiling

• Single-end

• Paired-end

• Multiplexing

p-value

• DESeq2

• edgeR

Correct fold change for variance

Often too few samples per group

1. treat each gene seperately -> estimate dispersion

2. fit a smoothing curve

3. shrink gene-specific dispersion towards expected dispersion

Visualization