pLM-BLAST

pLM-BLAST detects distant homologies between protein sequences by comparing ProtT5-derived single-sequence embeddings to identify homologous relationships, including local alignments, across low sequence identity.

Key Features:

• Distant homology detection: Identifies homologous relationships for proteins with substantial sequence divergence, including cases with less than 30% sequence identity.
• ProtT5 embeddings: Uses single-sequence representations (embeddings) generated by the ProtT5 protein language model for sequence comparison.
• Local alignment computation: Computes local alignments from embedding comparisons to detect partial or region-specific homology between proteins.
• Accuracy and speed: Demonstrates accuracy comparable to HHsearch while providing substantially faster performance for large-scale analyses.

Scientific Applications:

• Protein function prediction: Aids prediction of functions for uncharacterized proteins by identifying remote homologs with low sequence similarity.
• Evolutionary studies: Facilitates analysis of protein evolution by uncovering relationships between distantly related proteins.
• Protein annotation improvement: Enhances annotation of protein databases by revealing new homologous connections through local alignments of embeddings.

Methodology:

Compares ProtT5-generated single-sequence embeddings between input sequences and computes local alignments from these embedding comparisons.