Loading BioAtlas...
Loading BioAtlas...
Comparez modèles biologiques, moteurs de docking et flux de calcul côte à côte. Filtrez par modèle d'accès, empreinte de calcul et catégorie pour parvenir à une table de décision exploitable en moins d'une minute.
| Attribut | ＋ Ajouter un outil | |||||
|---|---|---|---|---|---|---|
| Category | Biological AI Models | Biological AI Models | Biological AI Models | Biological AI Models | Biological Foundation Models | — |
| Subcategory | Protein Design | Protein Design | Protein Design | Protein Design | Variant Effect Prediction | — |
| Company | Google DeepMind | Pacesa Lab | Boltz Team | Baker Lab | Meta AI (FAIR) | — |
| Access mode | Research Preview | Research code or publication | Research code or publication | Research code or publication | Open Source | — |
| Licence conditions | Research Publication / DeepMind Research | Check code and model weights separately | Check code and model weights separately | Check code and model weights separately | MIT License | — |
| Compute requirements | GPU cluster for generative diffusion sampling and candidate filtering | Check selected checkpoint and runtime; no GPU catalog recommendation | Check selected checkpoint and runtime; no GPU catalog recommendation | Check selected checkpoint and runtime; no GPU catalog recommendation | 1x GPU (>= 16GB VRAM) for batch scoring | — |
| Biological application | De novo binder design, targeted cancer therapeutics, viral inhibitors, and biosensors | Non enregistré | Non enregistré | Non enregistré | Predicting viral escape, disease-associated mutations, and enzyme thermostability | — |
| Research workflow | Specify target protein structure and epitope binding site -> generate de novo binder sequences and 3D backbones | Non enregistré | Non enregistré | Non enregistré | Wild-type sequence + candidate mutations -> compute log-likelihood ratio scores | — |
| Evidence level | DeepMind Research Whitepaper & arXiv preprint (2024) | Peer-reviewed primary method | Primary research preprint | Primary research preprint | Peer-reviewed research (NeurIPS 2021) | — |
| Commercial eligibility | Research publication & early academic collaboration | Unknown, requires audit | Unknown, requires audit | Unknown, requires audit | Permissive open source release | — |
| Price | Research Preview | Check selected model licence | Check selected model licence | Check selected model licence | Free Open Source | — |
| Availability ⓘ | DeepMind research publication and ongoing academic collaboration | Public research method/code | Public research method/code | Public research method/code | Available on GitHub and PyPI | — |
| Features | Generates high-affinity protein binders for diverse target proteins without initial starting fragments, Trained on large macromolecular structure repositories and AlphaFold predictions, Demonstrates up to 300-fold higher binding affinities than prior computational design methods, Experimental success rates ranging from 9% to 88% across diverse target proteins in wet-lab biophysical assays | AF2-guided optimization pipeline proposes protein binder designs for target structures., Candidate binder sequences, modeled protein complexes | Target-conditioned protein and peptide binder generation model as reported in its preprint., Binder sequence and structure candidates | All-atom protein structure diffusion conditioned on specific ligand or nucleic acid atomic context., Generated protein all-atom structures | State-of-the-art zero-shot variant effect prediction across 41 deep mutational scanning datasets, Ensemble of five distinct models trained on UniRef90 with different random seeds, Captures epistatic and structural constraints directly from evolutionary sequences, Standard benchmark baseline in computational biology for mutation scoring | — |
| Entry type | AI Model | Research AI model / method | Research AI model / method | Research AI model / method | AI Model | — |
| AI role | De Novo Binder Design | Published biomedical AI method | Published biomedical AI method | Published biomedical AI method | Variant Effect Prediction | — |
| Input data | Non enregistré | Target protein 3D coordinates; AF2/MPNN workflow and target interface constraints | Target context; design constraints | Target atomic geometry; designed region and constraints | Non enregistré | — |
| Output data | Non enregistré | Candidate binder sequences; modeled protein complexes | Binder sequence and structure candidates | Generated protein all-atom structures | Non enregistré | — |
| Validation | Non enregistré | Requires independent dataset-specific evaluation | Requires independent dataset-specific evaluation | Requires independent dataset-specific evaluation | Non enregistré | — |
| Type | Generative protein binder design model | Non enregistré | Non enregistré | Non enregistré | Masked protein language model ensemble | — |
| Intended use | Non enregistré | Non enregistré | Non enregistré | Non enregistré | Non enregistré | — |
| Compatibility | Non enregistré | Non enregistré | Non enregistré | Non enregistré | Non enregistré | — |
| Manufacturer | Google DeepMind | Non enregistré | Non enregistré | Non enregistré | Meta AI | — |
| Integration evidence | https://deepmind.google/discover/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/ | Non enregistré | Non enregistré | Non enregistré | https://github.com/facebookresearch/esm | — |
| Laboratory handoff | Designed binder sequences are ordered as synthetic DNA for yeast surface display and SPR/BLI affinity validation | Non enregistré | Non enregistré | Non enregistré | Guides directed evolution and deep mutational scanning library design | — |