Reconstruction of Ancestral Protein Sequences Using Autoregressive Generative Models.

Area of Science:

• Evolutionary biology
• Computational biology
• Biophysics

Background:

• Ancestral sequence reconstruction (ASR) aids understanding of protein evolution.
• Current ASR models often overlook epistasis, the context-dependence of mutations.
• Generative protein models have advanced, learning structural and functional constraints.

Purpose of the Study:

• To extend generative protein models for time-dependent sequence evolution incorporating epistasis.
• To improve the accuracy and reduce bias in inferring extinct ancestral protein sequences.

Main Methods:

• Developed a generative model capable of describing sequence evolution over time with epistasis.
• Applied the model to ancestral sequence reconstruction using protein families and evolutionary trees.
• Validated the method using simulations and experimental evolution data.

Main Results:

• The novel method outperforms existing state-of-the-art techniques in ASR.
• The approach enables sampling a wider diversity of potential ancestral sequences.
• This leads to a less biased characterization of ancestral protein states.

Conclusions:

• Incorporating epistasis into generative models significantly enhances ASR accuracy.
• The developed technique provides a more comprehensive tool for studying protein evolutionary history.
• This work advances our ability to reconstruct and understand past protein forms and functions.