To calculate the statistical quantities like variance or mean expression of a gene across cells, it is unfair to compare expression counts of a say gene A, if we know different cells has not be sequenced with sufficiently large net transcript count or library size and hence there is a low or high net resolution across cells due to variation in the library size. I am interested to know how does one deal with this problem. For example, in the attached plot, I have calculated the mean and std of every gene from a dataset across all the cells even though I show in the 2nd plot that the sum of the transcripts (library size) across cells have a distribution. I am sure because of this, certain highly varying genes maybe not having a high Fano factor or high variance when calculated where as low variance genes may be also picked up as highly variable. One way I did fix this issue is by looking at the statistics individually in each library size bin. Do you think there are some better methods out there yet ?
#scRNAseq #dropletsequencing #NGS #bigdatagenomics
Hello Sabyasachi,
as you said different library sizes should always be taken into account when analysing NGS data, especially for bulk- and sc-RNAseq data. There are several normalization methods that can be applied to scRNA-seq data in order to take into account the different library size issue. Since, NGS data are usually not normally distributed you cannot use statistical methods which assume a Gaussian distribution. For instance in this paper (Article Assessment of Single Cell RNA-Seq Normalization Methods
) the authors show that RUVr method is the best performing when you don't have spike-ins, conversely, the best method for datasets containing spike-in ERCCs is the gamma regression model (GRM).Then you can also decide to try to normalize the data respect to confunding factors, we use the software scater for this purpose.
I hope that this information might be useful.
Have a nice day
Riccardo
Bioinformatics data analyst at www.sequentiabiotech.com
Ricardo- thanks for the direction, I will definitely look up scater.
- Badges
- Science method