VL4-scRNA-Seq

• every cell is assayed

  -> 100% capture rate

• every transcript of every cell is detected

  -> 100% sensitivity

• every transcript is identified by its full length sequence

since 2017: in situ barcoding with microfluidic methods

• inDrop

• Drop-seq

• 10X

• At 4°C: bead releases barcodes

• At 50°C: cDNA synthesis with barcodes

• Droplet consists of cell and barcoding hydrogen bead in RT/lysis reagent, which are closed with oil

• Beads:

  • more than one

    -> too many barcodes -> only half or less transcript per bead

  • none

    -> no sequencing at all

• Cells:

  • more than one

    • different

      -> can’t tell which transcript comes from which cell

    • same

      -> too many (double) transcripts

  • none

    -> waste of resources, since only barcodes are then sequenced

= probability of one ore more beads in a droplet

= 1 - [probability of zero beads]

= 1 - e^(-μ)

= rate at which a captured single cell is associated with two or more beads

= caused by barcode collision

barcode collision = when two cells are separately encapsulated with bead having identical barcodes

barcode collision rate = expected proportion of assayed cells not having a unique barcode

-> can be avoided with a high relative barcode diversity

technical doublets = two or more cells in a droplet

technical doublet rate = given that at least one cell is in the droplet, the rate of having additional cells in the droplet

loading more cells, while having the same flow, increases technical and synthetic doublets

= two cells form a discrete unit, which does not break apart during disruption to form a suspension

-> avoid with nuclear scRNA-Seq

• raw data processing

• quality control

• normalization

• batch effect correction

• visualization

  • PCA

  • tSNE

  • UMAP

= Orthogonal linear transformation

• Pro:

  • Captures as much variance as possible

  • Good at showing global structure

• Con:

  • Poor at resolving local similarities

  • Sensitive to outliers

  • Not able to capture non-linear relations

However, non linear transformations perform better (tSNE, UMAP)

t-distributed Stochastic Neighbor Embedding

= unsupervised and non-linear dimensionality reduction technique, which preserves local similarities

Approach:

1. Select neighbors with Gaussian distribution over points in high dimensional space

2. Select neighbors with t-distribution ober points in low dimensional space

3. Minimize Kullback-Leibler divergence between both distribution

Uniform Manifold Approximation and Projection

= unsupervised and non-linear dimensionality reduction technique, which preserves local and global similarities

-> scales better than tSNE

Approach:

1. Approximate manifold for data in high dimensional space using simplical complexes as neighborhood graph

   -> K-simplex = convex hull of k+1 points (in k-d space)

   -> Consider radius around points to get overlaps, since overlaps can be represented by simplices and create simplical complex

   (overlap = x -> x-simplex)

2. Approximate distances for data in low dimensional space using spectral embedding

3. Optimize low dimensional fuzzy topology to be similar to high dimensional fuzzy topology via fuzzy set cross entropy

• meaningless cluster size

• meaningless distances between clusters

• misleading patterns

= downstream analysis at cell level using clustering

• useful for compositional analysis and cluster annotation

• clustering groups based on cells having similar gene expression profiles

  • simiarity based on distance metrics

    e.g. similarity scores from euclidean distances, calculated on PC reduced expression space

• Approaches to generate cell clusters from similarity score:

  • clustering algorithms

  • community detection methods

= downstream analysis at cell level with trajectory inference

-> molecular simulations to study the time evolution of a system's behavior over a series of discrete time steps

• useful for finding Metastable states and Gene expression dynamics

• interprets sc data as snapshots of a continuous process via trajectory inference methods

  • process is reconstructed by finding paths through cellular space with minimal transcriptional changes between neighboring cells

    • pseudotime variable = ordering cells along these paths = proxy for developmental time

      -> variable related to transcriptional distances from root cell

can be combined in coarse-grained graph representation, which represents both static and dynamic nature of data

-> sc clusters as nodes

-> trajectories between clusters as edges

= used to infer the direction and speed of changes in gene expression levels in single cells over time

• leverages the ratio of spliced and unspliced mRNA molecules to predict whether a gene's expression is increasing or decreasing in each cell

• providing insights into cellular dynamics and developmental processes

• helps to understand how individual cells are transitioning between different states

automatic annotation can be:

• marker gene database-based

• correlation-based

• supervised-classification-based