How to Use ProteinMPNN

ProteinMPNN is an inverse-folding model that designs amino-acid sequences from a known protein backbone. The Science paper showed that message-passing over backbone geometry can recover native-like sequences and generate experimentally viable designs across monomers and complexes, making the method a standard starting point for structure-based protein engineering.

On Neurosnap, researchers upload an Input Structure, optionally preserve key residues with Fixed Positions, and sample redesigns that remain compatible with the target fold. The workflow is useful for conservative scaffold redesign, interface-chain optimization, homo-oligomer sequence design, and rapid exploration of alternative sequences around an experimentally solved or predicted structure.

ProteinMPNN treats sequence design as a graph problem over backbone coordinates, learning residue probabilities from the local and long-range structural environment instead of from sequence homology alone. That fixed-backbone formulation is what makes it useful when a geometry is already trusted and the open question is which sequences can realize it.

On Neurosnap, Fixed Positions, Invert Selection, and Homo-oligomer define which parts of the structure are allowed to change and whether symmetry constraints should be respected. Number Sequences, Sampling Temperature, Model Type, and Model Version then control how broadly the model explores sequence space. The platform also exposes SolubleMPNN-style variants when solubility-aware redesign is preferable to a purely structural objective.

Researchers typically compare the returned sequence set rather than selecting one design blindly. Sequence statistics and amino-acid probability views help distinguish consensus-like redesigns from more exploratory proposals, which is useful before handing candidates to folding, stability scoring, or wet-lab screening.

• Structure-native design: ProteinMPNN starts from a concrete backbone, which is exactly the right input for scaffold-centered redesign.
• Precise region control: Fixed Positions, inverted selection, and homo-oligomer settings let researchers preserve catalytic or interface-critical residues while redesigning the rest.
• Sampling flexibility: Sequence count, temperature, and model-family choices support conservative recovery or broader sequence exploration.
• Experiment-ready triage: Probability maps and sequence summaries make it easier to shortlist candidates for folding, expression, and functional testing.