Likelihood scores

This page describes the different likelihood scores implemented in GeneRax.

The phylogenetic likelihood

The phylogenetic likelihood is the probability of observing a multiple sequence alignment given a gene tree. It is the score used by many probabilistic tools such as RAxML, IQTree and MrBayes. GeneRax supports all the models supported by raxml-ng.

It might sometimes be called "libpll" likelihood in GeneRax' logs.

The reconciliation likelihood

The reconciliation likelihood is the probability of observing a gene tree given a species tree. GeneRax implements:

• the UndatedDTL model that accounts for gene duplication, gene loss and horizontal gene transfers. See our paper for a more formal definition.
• the UndatedDL model, that is equivalent to the UndatedDTL model without HGT.

The joint likelihood

The joint likelihood is the product of the phylogenetic likelihood and the reconciliation likelihood. It is a powerful score that combines both models of sequence evolution and gene tree evolution.

In GeneRax and SpeciesRax

Here we describe which likelihood score is used in which step. Note that all the steps are optional and depend on the use case. Also, remember that the scores reported by GeneRax are log-likelihoods.

• the initial gene tree estimation from the gene sequences optimizes the phylogenetic likelihood (similarly to raxml-ng)
• SpeciesRax infers the rooted species tree that maximizes the reconciliation likelihood. It does NOT optimize the joint likelihood and does NOT use the gene alignments.
• the gene tree correction described in the first GeneRax paper searches for the gene trees that optimize the joint likelihood.

If you have any question regarding this page, please let us know on our google group!