retained rooted species tree when multiple possibilities

[Lulu84]

Dear David,

What determines which species tree root is kept for analysis when the following warning shows: " Multiple potential species tree roots were identified, only one will be analyzed"? (I have a

Is the "best option" selected somehow? If yes, is there a quantification or anything supporting a "rational" decision to keep the best rooted tree?

I am asking this because whatever test data I submit, the trees do follow perfectly the "true" or expected phylogeny.

Many thanks for your help! Lucienne

[davidemms]

Hi Lucienne

Yes, it applies an additional criteria to attempt to choose the best one. Thanks for asking about this, I'll give you the details here and then add it to the documentation.

It might help to refer to the STRIDE paper especially Figures 2 & 3 to help get a picture of what's going on. A. OrthoFinder applies the STRIDE method to find the root(s) which are in agreement with the largest number of well-supported gene duplication events (i.e. the parsimony method rather than the probabilistic model from the paper as this seemed to be the most accurate). If there is a tie, then you get the warning message you've observed and it applies a secondary method to find the 'best' one to use.

B. Choosing between multiple best roots according to STRIDE. We use two principles:

• There are a set of bipartitions in the species tree which have gene duplication evidence for their being a clade. We don't have enough evidence to be certain where in the tree the root should be, but wherever it is the rooting should make these biparitions into clades (rather than put a root in the middle of then), and ideally these clades should be placed far down in the tree (towards the leaves).
• As long as there we don't place the root the wrong side of any gene duplication events it doesn't actually matter if the root is in absolutely in the correct place, it will still be close enough to get all the orthologs right, and that's the most important thing for OrthoFinder.

Method: 1.For each potential root, find the topological distance from the potential root to the closest of these supported clades.

2. Select the potential root that is furthest from its respective closest supported clade. If there is more than one, pick the one with the largest branch length distance (i.e as a to resolve ties at the level of toplogical distance).

This ensures all supported clades (i.e. with evidence from gene duplications) are as far away as possible from the chosen root. This should minimise the possibility of any false positive or false negative ortholog assignments.

Updates to implement in OrthoFinder

• Document method
• Update the advice/warning given by OrthoFinder to the user in this situation. Steps user should take:
  1. Check the rooted species tree to see if it's correct.
  2. If not, re-run the final stage of the algorithm with the options "-ft RESULTS_DIR -s CORRECTLY_ROOTED_SPECIES_TREE"

All the best David