Unlocking Antibody Space With Databricks Vector Search

Antibodies form the bedrock of human health, and the body’s ability to create what is estimated to be a quintillion unique variants creates an impressive scaling issue for their study. Molecules derived from antibodies are now the fastest growing therapeutic category, and our understanding of how to design and modify antibodies requires solutions that scale. Databricks Vector Search provides an engineered solution for scaling up similarity search to 1 billion embeddings per endpoint, and by embedding antibody sequences with the protein language model AMPLIFY we will describe how this system might be used with the Observed Antibody Space dataset. The potential throughput of this system, at ~30 queries returned per second, would demonstrate state-of-the-art search. Using this scalable approach, we would then be positioned to explore advantages that this embedded representation affords for metadata lookup and statistical analysis.