Protein Design with Artificial Intelligence

De novo protein design seeks to find novel proteins which can carry out desired functions such as the binding of small molecules or the catalysis of chemical reactions. Traditional approaches rely on high-throughput screening methods that are time-consuming, expensive, and often unsuccessful. Artificial intelligence methods such as Large Language Models (LLMs) and Denoising Diffusion Probabilistic Models (DDPMs) have significantly transformed the field of de novo protein design and have had remarkable success in generating diverse proteins.

In our research group, we explore the capabilities of these new tools and apply them to the design of new-to-nature proteins. Currently, we are focusing on the design of new enzymes for plastic degradation using RFdiffusion and ProteinMPNN. Furthermore, to redesign existing proteins and achieve desired properties such as thermal stability or activity at different pH values, we develop deep learning methods for protein sequence design. This could help to expand the application range of enzymes to other chemical reaction conditions.

Selected Publications:

Collaborations: