straggling to combine multi-search outputs for global protein report

[asifratul]

Hi,

I am doing multi-omics (RNA and protein) experiments of 27 samples (3 Tissue x 3 time points x 3 replicates).
for proteins (Label-free) - I have already done searches using both Comet and MSFragger engines on HPC.
Now I want to combine both results together (within sample level), produce a combined_protein.tsv file with intensities and perform downstream differential expression analysis between samples.

As you suggested, I have made 27 folders for individual samples, each with 1 raw (centroid mzML) file, and 1 iprot.pep.xml file (combined comet and MSFragger pep.xml files using tpp iprophet). I can produce combined pep and prot files, followed by a protein.tsv file. but the protein file lack - individual spectral counts / individual sample intensities. I tried to initiate abacus, but it failed due to absence of "proper" pep.xml file! no luck with freequant as well !

Alternatively, I tried to keep both pep.xml files, so that - philosopher can combine both searches together. Even though pep.xml file generated, the program failed at protein prophet level!

Question:1 - can you please let me know, how to get a combined global output from iprophet inputs, with individual column intensities and spectral counts?

Question:2 - it might not be relevant, but if you can give some idea about the analysis pipeline, it would be much appreciated. for downstream stat analysis (MSStats / Limma-voom or similar R packages), can I use intensities from Freequant or Ion-quant would be better option? is there any normalisation step involved using "spectral counts"?

Thanks for your time.

ASIF

[prvst]

Hi @asifratul. Before I answer your questions, let me ask you a few things.

1. Are you using isobaric tags, like iTRAQ or TMT?
2. Why are you combining the output from two different search engines? What's your goal?

[prvst]

Yes, I thought so. Around that time, this kind of application was quite common, and if you search more, you'll find out that other groups published similar experiments where they showed how you can increase the identifications a little more by integrating different engines. With time, things changed quite a lot, and now this is not something you see people doing anymore (at least not for standard label-free DDA analysis). The main reason I would say is because the algorithms got a lot better, and the most popular tools became very similar in several aspects. If you feel that you can increase the number of identifications from your data, I would suggest alternative approaches, like using one of the "new generation" tools, like MSFragger, and doing open searches. In other words, I would suggest you start with one algorithm, and try to work with the search parameters to get the most of it, instead of adding other tools.

[asifratul]

The alternate approach you suggested, is it doing both open and closed searches and combining the files at peptide prophet?
can you give some insights about the downstream stat steps?

[prvst]

I would start by doing either an open search, or changing the database, including isoforms or proteins with lower confidence. The downstream would be very similar to the common approach because you dont necessarily need to identify the modifications, for example. I suggest you use FragPipe for that, and not just the Philosopher pipeline.

[asifratul]

I have trinity RNAseq assembly of same tissues. I have plan to use the predicted ORFs (~300k) as a db! Do you think the large size will effect the hits? (Larger db -> reduced sensitivity?)

sry not to be clear. I am familiar with both philosopher and fragpipe and already got the output files! After getting combined protein.tsv file, i want to do differential expression analysis. I didn’t understand difference between freequant and ion-quant intensities and which one would be suitable for the analysis. I have seen, msstats can produce nice matrics with intensity values, similar to rna seq count matrix! Can i simply use limma voom/ edgeR or similar bioconductor packages for differential protein analysis? Any normalisation step based on spectral count?

[prvst]

300 thousand entries would become a 600 thousand target-decoy database. That is definitively too much, so try to filter the low confidence ORFs from the database before doing the search. Freequant is based on MoFF algorithm, which takes the apex peak from the precursor, and IonQuant integrates the area under the curve. These are different approaches for doing MS1 quant. Both should work for you. Regarding the differential expression and the post-processing, there's a lot that one can do, so I suggest you check the May Institute Youtube videos and tutorials. They teach how to work with the data.

[asifratul]

Hi @prvst , Thanks for your suggestion. Previously, I used 3 engines (MSfragger, comet and MSGF+) for another exp. it involved metabolic N15 labelling of females and cross with "normal males", to find out the "nomal male" donated proteins in females.

I kept results of individual engines and samples separate and followed the philosopher pipeline.
I have found variations in their intensities (see attached list).

Do you think, it is happening due to "labelled" background?

[prvst]

It's hard to say without inspecting all the search parameters, and I don't think MSFragger supports the N15 searches.