constant-site correction interplay with proportion invariable (I) parameter

[Koen Vandelannoote]

I would appreciate some advice regarding the "proportion invariable" (I) parameter in the Site Model specification panel in BEAST2.

I typically run BEAST2 using a SNP alignment that contains only variable sites, as I work with bacterial genomes and including all sites would require a prohibitive amount of RAM. To account for the excluded constant sites, I manually add the constant-site correction to the XML following Remco's guidance in Correcting for constant sites in BEAST2. Correcting for constant sites in BEAST2

Conceptually, it seems reasonable to use a proportion-invariable model because the vast majority of sites in my bacterial genomes are invariant (e.g., >99.999%). My intuition is that I should estimate this parameter and perhaps provide a starting value such as 0.9.

However, in practice, BEAST2 only receives an alignment containing variable sites, with the constant sites represented indirectly through the constant-site correction. Given this setup, I am unsure whether it is appropriate to estimate the proportion-invariable parameter at all, or whether doing so would effectively double-count information already accounted for by the constant-site correction.

Could you advise on the recommended approach in this situation?

Cheers,
Koen Vdl