Sign in to access this content
Model
OpenFold predicts protein structures from protein sequence inputs and optional multiple sequence alignments (MSAs) and template(s).
Sign in to access this content
| Field | Response |
|---|---|
| Intended Application(s) & Domain(s): | Structural Biology for Drug Discovery |
| Model Type: | Pose Estimation |
| Intended Users: | This model is intended for bioscience researchers to infer protein structures from sequences for early-stage drug discovery and protein-related industrial processes. |
| Outputs: | Geometric Protein Structure (Text), Confidence Score (Optional), Embeddings (Optional) |
| Describe how the model works: | Predicts the pose of input protein sequence(s) from protein sequence alignments. |
| Technical Limitations: | This model does not work for nucleic acids and other small molecules. This model is less capable in predicting proteins that lack multiple sequence alignments (MSAs), such as complementarity-determining regions (CDR) loops in antibody and de novo proteins. |
| Verified to have met prescribed NVIDIA standards: | Yes |
| Performance Metrics: | 1. Coordinate deviation between experimental protein structure and prediction 2. Recovery of masked amino acid identity on protein sequence 3. Deviation between experiment and prediction distogram between pairs of amino acids 4. Uncertainty estimation by predicted Local Distance Difference Test (LDDT) |
| Potential Known Risks: | Model may inaccurately model biomolecular systems in drug development and bioengineering. |
| Licensing: | https://developer.download.nvidia.com/licenses/NVIDIA-BioNeMo-Framework-Evaluation-Software%20License(14Nov2023).pdf |