Fast, accurate construction of multiple sequence alignments from protein language embeddings

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Multiple sequence alignment (MSA) is crucial for understanding protein function and evolution.
• Traditional MSA methods struggle with low-sequence identity, often called the "twilight zone."
• Protein Language Models (PLMs) offer a new way to capture evolutionary and contextual information.

Purpose of the Study:

• To develop a new MSA algorithm leveraging PLM embeddings.
• To improve MSA accuracy and scalability, particularly in low-identity regimes.
• To demonstrate the effectiveness of PLMs in comparative sequence analysis.

Main Methods:

• Generated amino acid embeddings using PLMs.
• Developed a windowed reciprocal-weighted embedding similarity metric.
• Created the ARIES algorithm using PLM embeddings and dynamic time warping for global MSA construction.

Main Results:

• ARIES achieved higher accuracy than state-of-the-art methods on diverse benchmark datasets.
• The method showed particular strength in low-identity alignment scenarios.
• ARIES demonstrated near-linear scalability with an increasing number of sequences.

Conclusions:

• PLMs can significantly enhance MSA accuracy and scalability.
• ARIES represents a breakthrough in applying PLMs to comparative sequence analysis.
• This approach has the potential to transform the field of computational biology.