Clarification regarding use of maxLFQ for analysis of main report.

[vindr20]

Hi, I'm trying to transition from using the pg/pr_matrix files to using the main report, as seemes to be recommended in the documentation and several open 'Issues' on this github page. For the most part, I've figured out how to work with the report.tsv (I use fragpipe so I'm still on on 1.8.2_beta_8), but I am confused as to the correct way to integrate protein-level intensities with MaxLFQ.

I have made the diann_maxlfq() function (from the diann r package) and the fast maxlfq functions (out of the iq package) work, but they produce very different results from each other and from the pg_matrix that dia-nn outputs. I saw in another 'issue' that this is expected, as dia-nn uses a more sophisticated version of maxlfq than is implemented by either R package. However, dia-nn seems to produce (subjectively) better results than either R package does.

I'm curious what maxLFQ algorithm you recommend for working with the main report, and whether or not the more sophisticated dia-nn maxLFQ algorithm will ever be implemented in R (or another platform)?

[tvpham]

Hi,

Yes, there is a similar issue that is high on my todo list.

MaxLFQ as originally published is an optimization criteria rather than algorithm. The algorithms implemented in MaxQuant or DIANN might involve extra filterings and/or using different ways to deal with numerical stability. The iq algorithm uses 'graph connected components' to ensure the optimal solution.

I will get back to this later.

Cheers,
Thang