The overall strategy follows a logical three-step protein design pipeline: (1) generate a suitable backbone, (2) optimize its sequence, and (3) validate its structure.
Backbone Generation (RFDiffusion) RFDiffusion is used first because binder design fundamentally begins with identifying or generating a 3D scaffold capable of engaging the target epitope. A binder’s shape determines its ability to complement the epitope. Starting with generative backbone design allows exploration beyond natural protein folds, potentially discovering novel geometries better suited for high-affinity binding.
Sequence Optimization (ProteinMPNN / SolMPNN)
Once the backbone geometry is fixed, ProteinMPNN is used to determine the amino acid sequence that best fits that structure. A generated backbone alone is unstable without the correct sequence. ProteinMPNN efficiently searches the sequence space to find residue combinations that stabilise the fold and strengthen the target interface.
The final designed sequences are evaluated using AlphaFold2. Before experimental testing, it's crucial to verify that the designed sequence actually folds into the intended structure. AlphaFold2 serves as an in silico validation tool, confirming structural integrity and interface quality, and filtering out sequences likely to misfold.
No proteins found matching your criteria.