News

Antibodies are highly specific, avoiding off-target activities that challenge GPCR drug discovery. However, their pharmacological actions on GPCRs have been predominantly inhibitory. Abalone Bio’s Functional Antibody Selection Technology (FAST) platform combines synthetic biology and machine learning (ML) to discover and create functionally active antibody drugs—such as agonists, positive or negative modulators, or antagonists. FAST goes beyond mere binding by simultaneously measuring the function of 100 million antibodies against GPCRs, allowing us to find rare antibodies that are hard to discover through traditional screening methods with limited throughputs. FAST couples GPCR function to cell growth in a quantitative and tunable manner, enabling sensitive identification of functional antibodies with a wide range of activity. These discovery campaigns produce library-scale sequence-function datasets that uniquely power generative protein language models to design novel active antibody sequences.

‍

Using FAST, Abalone has identified multiple agonists against two GPCRs. FAST-discovered CB2 agonist antibodies have demonstrated activity both in vitro and in vivo. We constructed a sequence-to-functional activity map byfine-tuning a protein large language model on data generated as part of our discovery campaigns. We used the learned activity map to efficiently guide a generative model in sampling predicted active sequences from novel sequence space near the seed, then experimentally tested them. Our generative approach not only enriched for active novel sequences, but also created sequences with greater activity than the original sequence. To our knowledge, these are the first designed GPCR antibody sequences that are functionally active. Unlike other methods that generate stronger binders that may not translate to improved activity, the FAST platform uniquely trained on functional activity to create agonists with increased potency.