What is ProteinMPNN?
ProteinMPNN is a deep-learning model designed for protein sequence design given a backbone structure. It belongs to a class of "inverse folding" or "sequence design" tools.
Key Features
- It takes a 3D backbone (atomic coordinates of the protein scaffold) and predicts amino acid sequences that are likely to fold into that structure.
- It uses message‐passing neural network (MPNN) architecture to model interactions among residues.
- It is much faster than traditional physics-based methods (e.g. Rosetta) for designing sequences, while often achieving higher accuracy in terms of sequence recovery on native backbones.
- The tool has been validated experimentally across multiple systems: monomers, cyclic homo-oligomers, nanoparticles, binding proteins, etc.
How ProteinMPNN Works?
| Step |
Description |
| Backbone input |
A protein backbone structure (from experiment or predicted). |
| Encoding |
MPNN encodes information about backbone geometry and relations (distances/angles between residues). |
| Decoding / Sequence Prediction |
The model outputs a sequence that is likely to fold into that backbone. Some residues may be fixed (e.g., active site or binding site residues) to preserve function. |
| Filtering / Scoring / Validation |
Designed sequences are evaluated using structure prediction tools (AlphaFold2 etc.), metrics like folding confidence (e.g. pLDDT), RMSD, solubility etc. Then only top candidates are taken forward. |
Pharmaceutical / Biotech Applications
ProteinMPNN has been applied in a variety of practical contexts, relevant to pharma.
| Application |
Examples / Benefits |
| Stability / Expression Optimization |
Using ProteinMPNN to redesign native proteins (e.g. TEV protease, myoglobin) to improve thermal stability, expression yield, solubility, while retaining functional activity. |
| Rescuing Failed Designs |
Designing that failed with older methods (Rosetta etc.) were "rescued" using ProteinMPNN; designs folding correctly and showing binding etc. |
| Nanoparticle / Multimeric Assembly Design |
Designing two‐component protein nanomaterials (assemblies) more efficiently than Rosetta, with fewer computational resources, high success rates. |
| Functional Modulators / Binding Variants |
E.g., redesigning ubiquitin‐variants (UbVs) to modulate activity of the Rsp5 E3 ligase (enhancing its activity) through designed binders/variants. |
| Peptide PROTAC Design |
Designing binding peptides for AR and VHL as part of peptide PROTACs (targeted protein degradation agents) with downstream experimental validation. |
| Synthetic Binding Proteins |
Expanding sequence spaces of synthetic binding proteins (SBPs) to improve solubility, stability, binding energy relative to traditional engineering. |
Advantages
- Much faster design cycles; can generate many candidate sequences quickly.
- High sequence recovery and folding confidence, reducing wasted effort in screening non-folding or unstable proteins.
- Good at preserving functional regions when needed (fixing active site residues).
- Scalability: suitable for designing binders, stabilizing proteins, improving expression — many pharma R&D needs.
Limitations
- Input structure quality matters: errors in backbone or missing segments can reduce design quality significantly.
- Functional constraints (active/binding site behavior) may require fixing residues, which limits the redesign flexibility.
- Predictive validation (structure confidence, sequence recovery) still needs experimental validation to confirm activity, binding, stability etc.
- Not always optimal for large conformational changes or dynamic binding sites.
Related Services
Structure Modeling Service
Antibody-Antigen Interaction Modeling Service
Reverse Docking Service
Rigid Docking Service
Peptide Folding Simulation Service
* For Research Use Only.
Related Services
