Selecting the background for GO enrichment analysis (proteomics)

Question

[Sebastian Hesse]

Im using clusterProfiler for GO enrichment of differentially expressed proteins from different groups. It appears that most suitable is to use as a background for enrichment all proteins that "potentially could be in the lift of the differentially expressed proteins".

My issue is that I am not 100% sure what this means as I go though various rounds of selecting proteins before differential expression analysis. Eg I select only proteins quantified with >1 peptide.

Would you set the background for enrichment using really only the full list of proteins used in diff ex analysis (so the most stringently filtered ones), the list of proteins before filterin (so including proteons quantified with 1 peptide) or maybe even another list, eg the complete human proteome or the cells proteome as published by others before me?

Thanks a lot! Sebastian

ps: so far I am using the highly filtered list that includes only proteins used for diff ex analysis

[janetscully]

Oh my God!!! I've been doing enrichment with the whole proteome of Arabidopsis (I work with plants). Just to be sure: when it comes about annotation of gene IDs you DO use the whole genome info, right? (I am in a learning-by-doing process)

Thanks a lot !!

Janet

[Jean-Karim Heriche]

What you're asking is unclear. If you're asking about how to define the background/reference list for enrichment analysis, see my answer below. Otherwise, consider opening a new question.

Answer 1

[Jean-Karim Heriche]

The background should be the universe of all genes interrogated in the experiment, i.e. all genes that had a chance of making it into the list of genes of interest (e.g. differentially expressed ones). For example if your experiment only tested expression levels of transcription factors, then your background list would be limited to the transcription factors you tested.

[ATCG]

This doesn't make sense to me since we don't know the universe of TF. How could you filter your dataset( background) by categories for which we have limited knowledge?

[Jean-Karim Heriche]

The background is not the list of things of which you have knowledge, it's the list of what you tested/experimented on. If the experiment measured expression of 100 TFs then the background list for the purpose of enrichment analysis is the list of the 100 TFs you tested. It is impossible that any other TFs could be identified as interesting since they were not tested. Including them in the background creates a bias in the analysis.

EDIT: To try and clarify further, if your proteomics experiment started from the whole sample and your mass-spectrometer was set to identify all peptides then any protein in the sample had a chance to be detected. The filtering is applied as one of the steps to identify the proteins of interest.

[ATCG]

I see. Thanks for clarifying. How would you do this for an RNA-Seq experiment where you survey the whole transcriptome?

[Jean-Karim Heriche]

In this case, every transcript present in your sample has a chance of being detected (if no particular selection is applied to the RNAs such as polyA enrichment) so the background list should contain all the corresponding genes. Often we don't know what genes are actually expressed in a particular sample so we just take all the genes in the genome but sometimes we know to some extent which genes are not expressed in which case it would make sense to not include them.