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

ARIES builds PLM-based multiple sequence alignments that achieve higher accuracy than state-of-the-art methods, especially in low-identity regimes, while scaling almost linearly with sequence count.

• Why it matters: Traditional MSA scoring from conserved-family substitution matrices performs poorly in the low-identity “twilight zone,” limiting evolutionary and annotation analyses for distantly related proteins.
• What they did: The method uses a windowed reciprocal-weighted embedding similarity metric from protein language model embeddings, then aligns sequences to a PLM template via dynamic time warping to form a global MSA.
• The result: On diverse benchmark datasets, ARIES attains higher accuracies than existing approaches and remains efficient, enabling accurate, scalable MSA construction across protein families with varying sizes and similarity levels.

AI summary of the abstract · may contain errors · not medical advice · read the paper.