Fig. 1: Overview of the PLMSearch pipeline.
From: PLMSearch: Protein language model powers accurate and fast sequence search for remote homology
a PfamClan. Initially, PfamScan54 identifies the Pfam clan domains of the query protein sequences, which are depicted in different color blocks. Subsequently, PfamClan searches the target dataset for proteins sharing the same Pfam clan domain with the query proteins. Notably, the last query protein lacks any Pfam clan domain, and therefore, its all pairs with target proteins are retained. b Similarity prediction. The protein language model generates deep sequence embeddings for query and target proteins. Subsequently, SS-predictor predicts the similarity of all query-target pairs. c Search result. Finally, PLMSearch selects the similarity of the protein pairs pre-filtered by PfamClan, sorts these protein pairs based on their predicted similarity, and outputs the search results for each query protein separately. d PLMAlign. PLMAlign utilizes per-residue embeddings as input to compute a substitution matrix. This substitution matrix is then employed to replace the static substitution matrix in the Smith-Waterman (SW)49 or Needleman-Wunsch (NW)50 algorithm, enabling the local or global sequence alignment. The global alignment is illustrated in the figure, where the length of the query protein is 105, the length of the target protein is 123, and the embedding dimension of ProtT5-XL-UniRef50 used by PLMAlign is 1024.
