Selection Pressure Analysis

Question

[upasana]

Dear All,

Why Are My CodonML Results Different When Reproducing Published Work?

Recently, I attempted to reproduce the work from the following paper: Petersen, L., Bollback, J. P., Dimmic, M., Hubisz, M., & Nielsen, R. (2007). Genes under positive selection in Escherichia coli. Genome Research, 17(9), 1336–1343

In the paper, they identified several proteins under positive selection in Escherichia coli, including OmpC (Outer Membrane Porin C), which is one of the five beta-barrel porin loci showing evidence of positive selection. I followed the exact steps described in the paper using CodonML to reproduce their results, but I am seeing discrepancies in the values of omega and LRT (Likelihood Ratio Test). The paper used CodonML in PAML 3.14, whereas I am using CodonML in PAML 4.9.. I am curious to understand why there are these differences in results despite following the same methodology. Could it be related to differences between the versions of PAML (3.14 vs. 4.9j)? Does anyone have experience with this kind of discrepancy when using updated software versions for reproducing published results? Any insights into potential reasons would be greatly appreciated. Below is the comparison of my results with the published data:

Published Paper M7 vs M8 M1a vs M2a: LRT 40.49834 39.28 omega 58.08228 62.96

Current Analysis M7 vs M8 M1a vs M2a: LRT 48.05679 41.292804 omega 66.11662 67.22578

Thanks!

Answer 1

[DGTool]

It can likely be due to the updated version of the software. Not specific to this paper, but it is not uncommon for updated version of the software to have some slight (or major) changes to the way it might calculate things, or do some corrections etc. One example could be, for example, some proteomics software (e.g. spectranaut, DIA-NN etc) where a new version could mean you identify a higher number of proteins/peptides than before, meaning trying to recreate an old paper with a new version would not equal the same results.

If you do want see if you get the same results, I would suggest trying to use the same version as they had in the paper and see if that gets it closer.