The result of neutral.fit differs from other packages about miceco

[Russel88]

Hi Xiaoping,

The underlying problem is that R-squared is not appropriate to use for non-linear models, like the neutral model. (See for example this paper: https://bmcpharma.biomedcentral.com/articles/10.1186/1471-2210-10-6)

iCAMP and reltools use the standard approach to calculating R squared (1-SS_res/SS_tot), but that is only valid for linear regressions. In MicEco I use the generalised R squared. I think the MicEco version is more appropriate, but neither are probably "correct".

[lixiaopi1985]

That makes sense. Thank you, Russel!

Do you recommend rarefaction or any kind of transformation and normalization for amplicon data before fitting the neutral model?

[Russel88]

Yes, I think you need to rarefy, as the model uses the number of "Individuals" in each sample, which is assumed to be similar. The function (including that from reltools) just takes the mean read depth across samples as the number of individuals, and I am unsure how the model behaves if the read depths varies.

[lixiaopi1985]

Thank you so much!