Note: in the below video, the PDB file ‘1r42.pdb’ - which corresponds to the human ACE2 protein - was pre-loaded before the workflow’s launch. For a new design task, it will be necessary to repeat this step with a different .pdb file corresponding to the desired target protein.
It has recently been demonstrated that de-noising probabilistic diffusion models can generate realistic, or even life-like images in response to semantic prompting - which Watson et. al have adapted to the realm of protein design in their seminal work: RFDiffusion. This model has enabled the de novo generation of protein structures with pre-defined functional properties, specific to a chosen use case. By building on the structural understanding instilled in models like AlphaFold2 and RoseTTAFold, Watson et. al fine-tuned RoseTTAFold on a serious of protein structure denoising tasks to spectacular results.
The resulting model shows outstanding success in the prediction of accurate protein structures both in silico and in subsequent in vitro validation experiments. With proper constraints applied throughout the design loop, RFDiffusion is capable of executing on a diverse range of challenging design tasks, such as “unconditional and topology-constrained protein monomer design, protein binder design, symmetric oligomer design, enzyme active site scaffolding and symmetric motif scaffolding for therapeutic and metal-binding protein design”. The power and generalizability of the model make it a fearsome weapon for the bioengineer’s toolkit.
Read on for examples of how RFDiffusion can be used to design a binder for the human ACE2 protein.
Step 1: Navigate to RFDiffusion on Superbio.
Step 2: Click ‘About’ if you don’t already have a specific design task in mind. Superbio’s RFDiffusion implementation makes it easy to design proteins in one of the following categories: Unconditional Protein Generation, Motif Scaffolding, Binder Design, and Symmetric Oligomers Generation.
For this tutorial, we will experiment with binder design for the human ACE2 protein, best known as the binding partner of the RBD domain subunit of the Spike protein expressed by SARS-CoV-2.