Welcome to the E-score web server! Follow these simple steps to align two protein sequences using local, global, or semi-global alignment. The algorithms are based off the traditional dynamic programming methods (Smith-Waterman and Needleman-Wunsch), but with enhanced scoring functions. By utilizing a protein embedding model, each amino acid is represented as a numerical vector, capturing its biochemical properties and semantic meaning. These vectors are directly integrated into the scoring function during the alignment process in the form of cosine similarity. Please see the references below for more details.

Our testing identified Ankh as the optimal embedding model, consistently outperforming traditional BLOSUM scoring matrices. For user connivence we provide the following models: Ankh (default), ProtT5, ProstT5 and ESM2 here. Any embedding model can be used and we provide set up information for additional models (Ankh, ProstT5, ProtT5, ESM2, ProtBert, ProtAlbert, ESM1b, and ProtXLNet) on our GitHub page. Please note that user requests are processed in a queue, which may cause delays. For quicker results, using the source code directly is recommended. Users can provide their email addresses to receive results upon completion. Average wait times range from a few minutes to an hour, depending on protein length and server traffic. Once computed, results will be available for download and emailed if an address was provided. Please note, the email may be filtered to your junk folder.