That is not an easy one. Combining results from different database searches is not straightforward, as e.g. FDR calculation depends on the individual database background. 

Do literature search on "metaproteomics" - most is on multispecies environments from e.g. digestive tract, but your soil proteomics will present a similar problem.

It is indeed very difficult to compare/integrate results from different database searches. People generally choose score based or FDR filter to select results from individual databases and then simply compile the list of identification. Any conflicting spectrum (Cases where ranking would help) for which more that one peptide are assigned are generally ignored.

Alternatively, one can choose a database(Say Swissprot) as main search DB and append identified proteins' sequences from other database searches(NCBI etc.) in this database. Doing a new search in this compound database would give you single result file with PSMs properly ranked and Decoy criteria justified. 

In addition to what has been already said, I would like to point out the following.

A significant result on a curated database (such as swissprot) is much more reliable than one on a non curated database.

A huge, but redundant,  database will give you high scores for some entries simply because of the size of the DB.

A curated database has less, but more reliable,  entries. Numbers will be different, but the score and the chosen FDR will have much more confidence.

If your question is: "how do I rank a protein ID with a MASCOT score of 50 on NCBI and a score of 30 on SwissProt", I am afraid you will not find reliable methods to do so. 

As you are using Mascot as a search engine and the nature of the samples are not really from well annotated organisms, I would be more in favour to use a large and less well annotated database (NCBI) with more species and environmental genome. You can use the "Identity"  or "homology" threshold score which take into account the database size.

http://www.matrixscience.com/help/interpretation_help.html

http://onlinelibrary.wiley.com/doi/10.15252/msb.20145697/abstract

I would be incline to capture the overall average or median identity score (the higher this score is the less good is your database/data combination).

From a first list of hits you are generating, you can create a smaller database (try to keep it relatively broad).

good luck

T

Hello Gerry,

Did you analyze your samples using a high resolution instrument like an ltq-orbitrap / FT or Q exactive or bruker impact ?

If this is the case you can use MaxQuant. The software can handle different databases and FDR and identifications ranking can be done in different ways like score / p value / q value and so on.

thanks IIian

Analysis on LTQ-Orbi-trap

In this case you can give it a try. I hope it helps you with your question. Please consider all of the issues that were posted above because they completely / partially apply to this alternative analysis.

MaxQuant would aggregate redundant protein identifications together but you should be able to filter this based on some prior knowledge. There is a google group and wiki where you can get more information about it.

Note that MaxQuant will not solve the problem of comparing numbers from searches on different databases.

Reading again the initial question, I would say: do you actually need to rank these proteins all together?

I would present a table for each Database and add visual effect (background color) for proteins identified ion all of them etc etc.

iProphet tool from TransProteomic Pipeline can be used to combine different search outputs (even from different search engines such as x!tandem, comet and mascot) and to improve discrimination between true and wrong PSMs.

Maybe it helps.

Cheers

Hello again,

This tool came out recently (MetaProteomeAnalyzer):

http://pubs.acs.org/doi/abs/10.1021/pr501246w

It can parse mascot results.