1.4 Normalization Issues

Background

Problems and issues

• Sparsity of data and technical noise ("batch effects") --> will mask the signal of interest

Causes:

Spike-in for Normalization

RNA content (total amount and species) varies

for mRNAs:

• 92 ERCC molecules

• 8 mRNAs

• whole transcriptome HeLa RNAs

for sRNAs:

• 52(?) sRNA sequences

Caveat:

Typically only half of the spike-in were detected.

Computational Normalization Tools

for Single cell RNA-seq (and exRNA-seq)

1. scran:

   1. pools multiple cells (samples) in order to estimate cell-specific size factors in the presence of zero inflation and unbalanced differential expression of genes across groups of cells;

   2. precluster (using e.g. rank-based clustering) the cells into smaller, more homogeneous sets

2. SCnorm

3. Census

If considering spike-ins:

1. SAMstrt

2. GRM

References

• Normalizing single-cell RNA sequencing data: challenges and opportunities, Nature Methods, 2017

More Reading & Practice

See more about normalization, imputation and confounder (e.g. batch effect) in

• Additional Tutorial : 4.QC and Normalization; 5. Imputation and Confounders

Video

a) Normalization 1

@Youtube

@Bilibili

b) Normalization 2

@Youtube

@Bilibili