Ultrafast classical phylogenetic method beats large protein language models on variant effect prediction.

Area of Science:

• Evolutionary biology
• Computational biology
• Bioinformatics

Background:

• Amino acid substitution rate matrices are crucial for phylogenetics but computationally intensive to estimate.
• Current methods require reconstructed trees from massive protein alignments, creating a significant bottleneck.

Purpose of the Study:

• To develop a computationally efficient method for estimating amino acid substitution rate matrices directly from multiple sequence alignments (MSAs).
• To introduce a novel site-specific phylogenetic model (SiteRM) leveraging this efficient estimation.
• To evaluate the performance of SiteRM in variant effect prediction.

Main Methods:

• Developed a near-linear time algorithm, FastCherries, for cherry reconstruction from MSAs.
• Estimated rate matrices directly from MSAs, bypassing the need for pre-computed trees.
• Introduced the SiteRM model for site-specific rate matrix estimation.
• Applied SiteRM to variant effect prediction tasks using clinical and deep mutational scanning data.

Main Results:

• The FastCherries algorithm enables near-linear time estimation of rate matrices from MSAs, orders of magnitude faster than existing methods.
• The SiteRM model, utilizing site-specific rate matrices, demonstrates superior performance in variant effect prediction compared to large protein language models.
• The method successfully handles MSAs with millions of sequences, showcasing unprecedented scalability.

Conclusions:

• The developed method significantly accelerates the estimation of fundamental evolutionary parameters.
• SiteRM offers a powerful new tool for variant effect prediction, outperforming existing state-of-the-art models.
• This work advances statistical phylogenetics and computational variant effect prediction through efficient algorithms and novel modeling approaches.