Motif scaffolding is one promising direction in computational protein design where de novo protein scaffolds are designed to maintain a set of functional protein coordinates (a motif). Here we explore an AF2-Multimer hallucination workflow, MotifCraft, for motif scaffolding and present approaches that improve efficiency while maintaining or improving binder generation accuracy.

MotifCraft shows that AF2-based hallucination can be used for motif scaffolding by initializing with the motif coordinates and backpropagating through the AF2 network using motif-specific losses. This novel method produces higher in silico success rates than existing diffusion-based scaffolding methods. Furthermore, we show that scaffolding a binder interface motif results in designs scoring better according to known in silico interface metrics (ipSAE, ipTM, RMSD) when we include the target protein in the workflow.

Finally, we compare two methods for accelerating binder design through AF2-backpropagation. In the first, we crop the target protein to the minimal binding interface before passing through the network. For the second, we allow the full target protein to go through the AlphaFold2 Embedder module and then crop before the more computationally heavy Evoformer module. We conclude that cropping the target protein is an effective method for scalable binder design against large proteins. Notably, both implementations provide slight advantages, depending on the target protein.

MotifCraft thus enables fast, accurate, and scalable motif scaffolding as well as binder design against large protein targets that were previously inaccessible to AF2-hallucination-based methods.

Co-authors: Torsten Schwede, Janani Durairaj