how to fix low RNA input in bulk RNAseq data analysis?

Question

[Sara]

I have some RNAseq data and when I got count data, I checked the expression of some house keeping genes and in few samples I saw they are up to 10 fold less than other samples showing that RNA input was very low in those samples. what should I do with those samples? shall I exclude them from the analysis or there is a way to fix this?

Answer 1

[Trivas]

There are a couple ways of looking at it:

1. Do your samples still correlate with biological/technical replicates (e.g., correlation matrix)?
2. Do your samples cluster similarly in a PCA plot?
3. Does your upstream QC suggest that low input or low quality RNA was used?

Finally, I've removed samples when the normalized gene counts are outliers from the rest of the samples using a boxplot of total normalized counts.

[ATpoint]

That depends on how samples are distributed with regard to experimental groups. If in the worst all controls are undersequenced and all treatments are not then there is not much you can do. Maybe remove genes that consistently have low counts in controls.

If it is somewhat balanced between groups you might use voomWithQualityWeigts() or arrayWeights() from limma to downweight outliers rather than hard-filtering them. Can you add any details or diagnostic plots as suggested by Trivas?