"De novo deisgn of protein and chemical interactions using deep learning"

ABSTRACT

The breakthrough in protein structural biology using Artificial Intelligence (AI) transformed protein structure prediction and protein design. However, problems that require atomic-level accuracy and high resolution modeling such as small-molecule binding protein design and enzyme design still remain challenging. We tackled this problem by developing a new protein design computational method based on the state-of-the-art AI models. RFdiffusion All-Atom is a generative AI model trained using RoseTTAFold All-Atom. RoseTTAFold all-atom can predict complex structures of all types of life’s molecules including proteins, chemical modifications, nucleic acids, metal ions, and small-molecules. The generative AI model RFdiffusion All-Atom was used to generate customized backbone structures for the target small-molecules. LigandMPNN enables generating sequences with given protein backbone and bound small-molecule atomic coordinates. By combining these methods with the Rosetta biomolecular software, we developed a next generation protein design platform that considers protein side chains or small-molecules in all-atom level. Using the new method we designed binders for eight different small-molecule targets including metabolites and small-molecule drugs. The binders were readily expressed in E. coli, with binding affinities ranging from nanomolar to low micromolar, straight out from the computer. The design structure accuracy was confirmed with protein and ligand co-crystal structures of the cortisol binder and the apixaban binder being in very close agreement with the design models. The success in designing interactions between proteins and chemical groups now allows us to pursue developing de novo protein-based tools that can be used to study and modulate biological systems in very high resolution and accuracy. We are actively extending our methods for de novo enzyme design, sensor design, synthetic biology, and the study of protein and nucleic acid modifications in epigenetics.

*The seminar will be held both offline and online via Zoom  (https://kaist.zoom.us/j/88244379581)