Unmet Need
Monoclonal antibody therapies have become an extremely popular and effective treatment route for cancer, rheumatoid arthritis, and infectious diseases. At present, there are over 100 FDA approved monoclonal antibody therapies with many more in late stage clinical trials (see Nature). One approach that is commonly used to generate monoclonal antibodies for therapeutic use is to screen a large library of randomly generated antibodies for specificity to a given target. However, this process produces large amounts of non-functional antibodies and many that successfully bind to the target antigen face problems with large scale production due to poor stability or low expression rates (see UpToDate). The generation of antibody libraries can be improved by better understanding sequences in antibodies that lead to better stability, safety, and solubility and prescreening the library sequences to preferentially include the most promising candidates. Therefore, there is a strong need for technology that improves antibody library design to be developed to increase the efficiency of therapeutic antibody discovery.
Technology Overview
Johns Hopkins researchers have developed Immunoglobulin Language Model (IgLM), a generative language model that can generate full antibody sequences or redesign targeted segments to generate an antibody library that produces antibodies with biophysical and immunogenic properties that are favorable for commercial antibody production. IgLM was developed by training on 558 million antibody heavy and light-chain variable sequences and uses bidirectional context for either de novo design or optimization of sequence spans of varying lengths. Altogether, IgLM allows for the production of antibody libraries with improved biophysical properties and lower immunogenicity compared to baseline design.
Stage of Development
Deployable model freely available for non-commercial use at https://github.com/Graylab/IgLM.
Commercial use license available at https://jhtv.e-lucid.com/product/generative-language-model-for-antibody-sequence-design-iglm.
Publication
Shuai, R. W., Ruffolo, J. A., & Gray, J. J. (2021). Generative language modeling for antibody design. bioRxiv. https://doi.org/10.1101/2021.12.13.472419
