Origins and evolutionary diversification of the protein-fold universe

Determine how life reuses successful structural solutions across vastly different species and establish how the universe of protein folds first arose and diversified over evolutionary trajectories.

Background

The conclusion places biologically informed tokenization within a wider set of unresolved questions in evolutionary biology. In particular, the authors identify two explicit open questions: the mechanisms by which evolution repeatedly reuses successful structural solutions across species, and the historical origin and diversification of the universe of protein folds. The LEMON results do not resolve either question; they are cited as examples of broader scientific problems that may benefit from improved sequence representations.

References

The advantage of this approach extends well beyond benchmarks: remote homology detection, protein engineering, and some of the deepest open questions in evolutionary biology, including how life reuses successful structural solutions across vastly different species and, more fundamentally, how the universe of protein folds first arose and diversified over evolutionary trajectories.

— LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling  (2609.37675 - Banerjee et al., 29 Sep 2026) in Section 6, Conclusions