Assessing phylogenetic inference in the era of AI-driven protein structures

Nikolai Romashchenko, Aurelie Fanchette, Dongwook Kim

Phylogenetic inference is one of the most common downstream applications of multiple sequence alignment. Quantitative assessment of inferred trees is essential for evaluating the reliability of both alignments and phylogenetic inferences. Establishing a robust assessment framework for phylogenetic trees therefore provides a basis for examining and improving methods for alignment and tree inference. In sequence-based phylogenetics, decades of development have produced a wide range of evolutionary models, support measures, and benchmarking strategies. Recent advances in AI-based protein structure prediction are now extending phylogenetic analysis to structural data at an unprecedented scale. This shift creates new opportunities to investigate deep evolutionary relationships that may be obscured at the sequence level. However, many established assessment methods rely on assumptions developed for molecular sequences and cannot be transferred directly to structural representations. In this review, we survey established methods for phylogenomic assessment and reconsider how they can be applied to the current state of structural phylogenetics. We aim to provide guidance for evaluating phylogenetic inference in the era of AI-based protein structures, facilitating the methodological advances towards robust standards for structural phylogeny.