Choosing the number of genes for species tree

[edwhisnant]

Hi,

I am curious if there is a quantifiable way of determining the number of genes one will ultimately choose to filter and use for the gene tree & species tree construction?

I really like the way phyling is able to determine "informativeness" of the genes using treeness/rcv. However, when deciding how many genes to retain, do you have a way of choosing the number of genes to use?

What I have done previously was I looked at the RF distance between resulting species trees, using an increasing number of filtered genes for the construction (up to the max w/ no filtering). Then I compare each tree to the "maximum tree" to see where RF distance plateau's with increasing number of genes used.

Do you suggest a better way to approach this?

[chtsai0105]

Hi @edwhisnant,

Sorry for the late reply! I’ve been focused on my dissertation over the past few months and didn’t realize a thread had been created.

Currently, we use an arbitrary number of markers specified by the user and don’t yet have an automated method to determine the optimal number. In most cases, even a relatively small subset - sometimes fewer than 50 markers - is sufficient to resolve the phylogeny. One of the key features we emphasize is speed, and using fewer markers can significantly reduce computation time during downstream tree inference, especially when running more sophisticated methods like IQ-TREE or RAxML.

The RF distance approach you mentioned is an excellent idea! Thank you for suggesting it. I'll run additional tests to estimate a suitable default number of markers and will likely incorporate this into the filter module.