This is a test version of Biostars. For the public version, visit https://www.biostars.org.
No significant genes after FDR correction in microarray (n=3) – how to proceed with exploratory analysis?

Hi Everyone,

I am analysing a two colour Cy3 Cy5 microarray dataset using limma. I have 3 biological replicates per condition.

I observe several genes with high log2 fold change and raw p values below 0.05. However, after multiple testing correction, no genes pass FDR threshold. The minimum adjusted p value is approximately 0.12.

My goal is exploratory pathway level analysis rather than reporting definitive differentially expressed genes. I am interested in performing GSEA or ORA to identify biological processes that differ between conditions.

I would appreciate advice on the following.

Is it acceptable to use raw p value and fold change thresholds for exploratory pathway analysis in such cases.

Would ranking genes using moderated t statistics or other ranking metrics be more appropriate for GSEA without applying strict DEG thresholds.

Are there recommended best practices for handling small sample microarray datasets where FDR correction removes all genes.

Thank you very much for your guidance.

microarray limma
• 914 views • link
updated by Ram Ram • written by Samaksh • Samaksh

1 answer

Would ranking genes using moderated t statistics or other ranking metrics be more appropriate for GSEA without applying strict DEG thresholds.

That is the whole point of GSEA. Taking all genes and rank it. There is no strict DEGs in a competitive geneset test, by definition.

• 0 views • link
written by ATpoint ATpoint

Hi ATpoint

Apologies for the confusion created in my previous question. I would like to clarify what I am trying to do.

I am working with a two colour Cy3 Cy5 microarray dataset analysed using limma with n=3 per condition. After multiple testing correction, no genes pass FDR threshold, although several genes show noticeable log2 fold changes and raw p values below 0.05.

My goal is exploratory comparison across three conditions where I want to extract upregulated and downregulated gene lists and perform Venn overlap analysis before downstream pathway interpretation.

Since FDR filtering yields zero genes, I am considering defining gene sets using nominal p value and fold change thresholds, for example log2FC > 1 and p value <= 0.05 for upregulated genes and log2FC < 1 and p value < 0.05 for downregulated genes.

Would this be considered reasonable for exploratory overlap analysis in small sample microarray datasets, provided the results are interpreted cautiously?

Alternatively, is there a more recommended strategy for generating gene lists for overlap comparison when FDR filtering is too stringent?

Thank you for your guidance.

• 0 views • link
written by Samaksh • Samaksh

You can do GSEA. One major usecase is indeed avoiding strict per-gene cutoffs. You can do anything you want to build a hypothesis. Just do and see whether it gives you a hint to follow up. 3 vs 3 in arrays is low, I would not even have done such an experiment unless the expected effect is super strong, like cancer vs normal. Just explore in different ways and follow up on what seems interesting and novel.

• 0 views • link
written by ATpoint ATpoint

Log in to answer this question.

More posts like this