Clustering 3K PBMCs with Scanpy

Contributors

Mehmet Tekman

Questions

• What are the main steps of scRNA-seq?

• What kind of variation can confound an analysis?

Objectives

• Learn the main stages of an scRNA-seq analysis

• Understand the methods and concepts underlying scRNA-seq

Single Cell RNA Pre-processing

Speaker Notes

1. Barcode Extraction
2. Mapping
3. Gene Annotation
4. Batch count matrices

Single Cell RNA Downstream Analysis

Speaker Notes

1. Filtering
2. Normalising
   • Confounder Removal
3. Dimension Reduction
4. Clustering
   • Annotation

Barcoding Cells

Filtering: Cell and Gene

Normalisation: Technical Variation

Normalisation: Biological Variation

Dimension Reduction: Relatedness of Cells

Speaker Notes Build a KNN graph from distance matrix:

• If P and q share distance which is

Dimension Reduction: Projection

Speaker Notes

• Can use tSNE, PCA, UMAP

Community Clustering: Louvain

Aim: Maximise internal links and minimise external links

Community Clustering: Louvain

Pick a cell, place in neighbour, and accept if internal:external increases

Cell Type: Identifying Cluster Types

.left-column50[.image-100[]] .left-column50[.image-100[]]

Clustering: Hard vs Soft

.pull-left[

.image-100[]

Hard

• Big spaces between clusters
• Cell types are well defined and the clustering reflects that

]

–

.pull-right[ .image-100[]

Soft

• Clusters bleed into one another
• Cell types seem to intermingle with one another.

]

Speaker Notes Why? Why would there be clusters so close to one another?

Continuous Phenotypes:

.image-100[]

Speaker Notes

• Cells aren’t discrete, they transition

• Continuously changing over time from a less mature type to more mature type

Interactive Environments: live.usegalaxy.eu

Speaker Notes What is Differential Expression in scRNA-seq?

CellxGene Local Test

• We can probe clusters to see how they are so differentially expressed

    pip3 install cellxgene
    cellxgene launch https://cellxgene-example-data.czi.technology/pbmc3k.h5ad
  

• Launch locally: http://127.0.0.1:5005

Key Points

• Understanding the purpose of barcoding
• Knowing the difference between hard and soft clustering
• A KNN graph can be generated from a count matrix.
• Community clustering can be generated from a KNN graph.
• Interpreting scRNA-seq plots

Thank you!