Bioinformatics Seminar
Time:
Venue: Na
10 September 2019
NaSample weights for count based differential expression analysis
Luke GandolfoWEHI Bioinformatics
When analysing "expression" data in the form of counts (e.g. RNA-seq ; ATAC-seq ; ChIP-seq ; etc.) outlier samples are sometimes encountered ; i.e. samples that fail to cluster properly in a PCA plot. These outlier samples can be problematic since they can significantly reduce our ability to detect differential expression. Instead of dropping such samples from the analysis ; one approach to dealing with them is to use sample weights: retain the samples in the analysis but down-weight their influence ; where the required amount of down-weighting is estimated directly from the data. Methods do exist for estimating sample weights ; but they are designed to be used in analyses where the count data has been log transformed (e.g. limma/voom analyses). This talk will present a new approach for estimating sample weights for analyses that work directly with the count data (e.g. edgeR analyses).;