Removing low level expression genes from microarray experiments

Question

[K.patel5]

Hello biostars,

I have pre-processed mRNA data from an illumina whole-genome expression array Human HT-12 V4 and microRNA data from an Agilent GeneSpring GX 11 array.

I am aware that, low level expression genes (<10) should be removed from RNA-seq analysis. Is there such a threshold or requirement to do so with microarray data.

I have not come across any literature that says so, but I would like to know if others remove low level genes from microarray data.

Any help or advice is appreciated.

Answer 1

[h.mon]

In general, microarray probes should be filtered if their intensity is not above the background intensity - how to detect this varies by array platform.

Check the limma User Guide, it shows how to filter "non-expressed" probes in several cases. For example, for the Illumina BeadChip Arrays - like the Human HT-12 V4 - the User Guide says:

Filter out probes that are not expressed. We keep probes that are expressed in at least three arrays according to a detectionp-values of 5%:

expressed <- rowSums(y$other$Detection < 0.05) >= 3
y <- y[expressed,]
  

See page 111 of the User Guide (revision of 23 January 2019), read both 17.3.4 How many probes are truly expressed? and 17.3.5 Normalization and filtering.

If you read this guide carefully, it will show how to filter noisy probes for other types of arrays as well.

Answer 2

[ATpoint]

The only thing you can do is to remove probes with poor intensity values which might be an indication for hybridization problems. A histogram might help you decide if there is a visual cutoff for filtering. Microarrays only give you relative expression values, therefore you cannot filter by expression values.