Another crosslinking and database partitioning issue?

[emirzakh]

Hi, I've noticed that when you adjust the database partition size, the number of identified inter and intra-protein crosslinks, above 1% FDR, increase 5-10 fold. My protein databases are fairly small, between 100-250 proteins, depending on the search. When using a database partition size of 1, the results are lackluster. Increasing the partition size to 2 brings the results closer in line with pLink2 and XlinkX. Increasing the partition size further doesn't have a significant impact on results when compared to a partition size of 2, until you get up to higher partitions like 25-50, which then results in a drop in crosslinks identified above 1% FDR.

For example, with an unenriched crosslink file, with all other parameters held constant (parameters aren't very stringent - 20 ppm precursor and product tolerance, 2 miss cleavages allowed, 2 variable mods, etc) I see the following:

Database partition - Inter - Intra - single - loop - deadend 1 0 2 10150 72 822 2 418 48 10067 77 814 3 295 57 10094 79 819 5 229 59 10112 81 836 25 70 62 10208 107 845

With an enriched crosslink sample, the results are also very dramatic:

Database partition - Inter - Intra - single - loop - deadend 1 8 846 101 781 2612 2 1092 1300 91 803 2622

I appreciate any advise as to whether the results achieved with the increased partition size can be trusted based on q-value/score or if there is a bug that is artificially causing high scoring crosslinked peptide ids.

[lonelu]

Thanks for providing the details of the issue. We are aware of the issue (#2039). We have a new update here which will solve this problem (#2084) in theory. Still, your information is very valuable and I need to run some analysis to confirm it. I didn't expect the difference of database partition to cause such a big change of ids. Please wait for more information and the update.

[acesnik]

Solved in https://github.com/smith-chem-wisc/MetaMorpheus/pull/2084?

[lonelu]

Should be. But need further feedbacks.

[emirzakh]

The recent MM update has affected the crosslink IDs again. Prior to this update, when searched with the database partition=2, a sample set gave 2,936 inter-protein, 953 intra-protein, 444 loop, 4689 mono, and 52,548 single peptides. I reran this exact search on the updated MM and the difference is very dramatic for the inter and intra-protein crosslinks - 13 inter and 25 intra. Loop, mono, and single peptides are still within similar ranges with 324, 3810, 40,751 respectively. Changing the database partition no longer rescues the results.

[trishorts]

to clarify:

1. you get a different result now compared to earlier
2. now, when you change partitions, you get the same result each time. (I think this is the desired result, correct?)

[emirzakh]

Hi, yes, the results now are different than prior to the update.

Regarding question #2 - yes, the result stays the same when the partition is changed. However, this raises another question - were the prior results correct or are the current results correct or were they both correct? With the massive difference of 3,889 vs 38 inter-peptide crosslinks for just 1 experiment between the two updates, this seems like an important answer to know.

[lonelu]

The results of our test case are also different from the results obtained with the previous version. However, they are not very different like yours. I have no idea what happened. Do you mind sharing part of your data for us to analyze? It could help us to figure out a potential bug.

[emirzakh]

Sure. Is there an email that I can send a google drive link to?

[lonelu]

Thank you very much! Please email to '[email protected]'. Please also include the .toml files you used for the current and previous versions.

[emirzakh]

Hi, I just wanted to double check that you were able to access the google drive files I sent. Please let me know if there are any other files that I can share to help with this issue!