Loading BioAtlas...
Loading BioAtlas...
Compare biological models, docking engines, and computational workflows side-by-side. Filter by access model, compute footprint, and category to reach an actionable decision table in under a minute.
| Attribute | ＋ Add tool | |||
|---|---|---|---|---|
| Category | Drug Discovery & Chemistry | AI Drug Discovery | AI Drug Discovery | — |
| Subcategory | Molecular Foundation Models | Research AI Methods | Molecular Property Prediction | — |
| Company | IBM Research | Research / method authors | Luo Lab | — |
| Access mode | Open Source | Confirm distribution and exact release | Research code or publication | — |
| Licence conditions | MIT License | NOT_INDEPENDENTLY_AUDITED | Check code and model weights separately | — |
| Compute requirements | 1x GPU or CPU for inference; multi-GPU for fine-tuning | Version and data dependent | Check selected checkpoint and runtime; no GPU catalog recommendation | — |
| Biological application | Solubility, toxicity (hERG, Ames), BBB permeability, and binding affinity prediction | Not recorded | Not recorded | — |
| Research workflow | SMILES input -> high-dimensional latent molecular embeddings -> downstream property predictions | Not recorded | Not recorded | — |
| Evidence level | Peer-reviewed research (Nature Machine Intelligence 2022) | PEER_REVIEWED_METHOD_OR_CONFERENCE | Peer-reviewed primary method | — |
| Commercial eligibility | Open source release | NOT_INDEPENDENTLY_AUDITED | Unknown, requires audit | — |
| Price | Free Open Source | Verify access and licence | Check selected model licence | — |
| Availability ⓘ | Available on Hugging Face and GitHub | Method paper or official documentation | Public research method/code | — |
| Features | Trained on 1.1 billion SMILES strings from PubChem and ZINC databases, Linear attention mechanism enabling efficient scaling to large molecular graphs, State-of-the-art benchmarks on MoleculeNet across physical chemistry and biophysics tasks, Fine-tuning support for custom pharmaceutical assay endpoints | ADMET-AI scientific method, Molecular SMILES with well-defined compound stereochemistry and relevant assay domain, Predicted absorption, distribution, metabolism, excretion and toxicity endpoints | Chemical foundation model for learned molecular representations, property prediction and conditional molecule generation., Molecular features, predicted properties or generated SMILES | — |
| Entry type | AI Model | Source-scoped research method | Research AI model / method | — |
| AI role | Chemical Foundation Model | ENSEMBLE_ML | Published biomedical AI method | — |
| Input data | Not recorded | Molecular SMILES with well-defined compound stereochemistry and relevant assay domain | Valid SMILES; trained downstream endpoint or conditioned molecule task | — |
| Output data | Not recorded | Predicted absorption, distribution, metabolism, excretion and toxicity endpoints | Molecular features; predicted properties or generated SMILES | — |
| Validation | Not recorded | Published method scope; no independent BioAtlas execution or benchmark | Requires independent dataset-specific evaluation | — |
| Type | Chemical transformer foundation model | Not recorded | Not recorded | — |
| Intended use | Not recorded | Not recorded | Not recorded | — |
| Compatibility | Not recorded | Not recorded | Not recorded | — |
| Manufacturer | IBM Research | Not recorded | Not recorded | — |
| Integration evidence | https://github.com/IBM/molformer | Not recorded | Not recorded | — |
| Laboratory handoff | Filters high-risk toxic molecules before compound synthesis | Not recorded | Not recorded | — |