RFdiffusion All-Atom is a deep learning framework designed to generate biomolecular structures, particularly proteins, around small molecules, metals, etc. It builds proteins de novo by modeling atomic interactions to create highly specific binding pockets. By learning atomic-level interactions between molecules and proteins, RFdiffusionAA generates new proteins with specific structural solutions, which can be validated through in silico and experimental means. LigandMPNN can be used with RFdiffusion-AllAtom results to refine designed protein sequences and optimize interactions with ligands for enhanced binding affinity and specificity.

Example use case: Designing custom proteins with specific small molecules

Technology: Diffusion

Limitation:

• Some of the parameters were kept default. Please see this page for more details.
• Accuracy depends on training data size and diversity, especially for novel biomolecular interactions not well-represented in current datasets. Experimental validation remains essential for designs.

Example parameter inputs:

• Ligand Binder:

Contig Map Input: 150-150

Ligand Name: OQO

• Ligand Binder with Protein Motif:

Contig Map Input: 10-120,A84-87,10-120

Ligand Name: CYC