Our vision is to make functional protein design as intuitive and programmable as writing Python code. We leverage cutting-edge generative AI to establish a programmable, controllable, and modular methodology for de novo protein design. Using these algorithms, we engineer proteins that can be precisely controlled by external stimuli such as pH, temperature, light, ions, and small molecules. Furthermore, we apply our algorithms to develop novel protein therapeutics, including antibodies, nanobodies, peptides, and miniprotein drugs.

We aim to fuse advanced AI models with high-throughput systems in wet-lab to accelerate the development of TCR immunotherapies for cancers. First, we are building foundational AI models to decode the complex rules of TCR recognition of antigens. Using these models, we identify potent, natural T cells specific to targets of interest like neoantigens. Second , we leverage generative AI to design de novo binders targeting pMHC complexes, expanding the toolkit for personalized immunotherapies.

We believe we are only witnessing the dawn of Generative AI, with significant challenges and opportunities remaining. Our research focuses on advancing diffusion-based generative models for both continuous data like images and 3D molecular structures and discrete data, including natural language and biological sequences. Furthermore, we are dedicated to optimizing model alignment using reinforcement learning to ensure safety and efficacy.