Protein similarity search using ProtT5-XL-UniRef50 and Amazon OpenSearch Service

This blog post describes a solution for performing similarity searches on protein sequences using embeddings generated by the ProtT5-XL-UniRef50 model and stored in Amazon OpenSearch Service as a vector database.

Specifically, the article covers:

• An overview of the solution architecture, which includes deploying the ProtT5-XL-UniRef50 model on Amazon SageMaker to generate protein embeddings and storing pre-computed embeddings for 20,000 proteins in an OpenSearch Service index.
• An explanation of embeddings and their importance for representing complex data like proteins in a vector space, enabling efficient similarity searches.
• A demonstration of the solution's capabilities by performing a similarity search for the Immunoglobulin Heavy Diversity 2/OR15-2A protein, which successfully retrieved proteins from the immunoglobulin family with similar bio-functional properties.
• Potential enhancements, such as tuning the OpenSearch Service KNN algorithm and scaling by importing additional protein embeddings.
• Conclusion highlighting the solution's effectiveness in calculating protein embeddings and finding similar proteins, with a link to the GitHub repository containing the code.