Pauling runs validated physics-based tools (UniDock, GROMACS, P2Rank) -- real simulations you can audit end-to-end, on a self-serve platform anyone can use today. Iambic's NeuralPLexer claims better structure prediction than AlphaFold 3 and their Enchant model predicts clinical endpoints. Impressive ML -- but they're surrogates inside an internal pharma pipeline that isn't publicly available. When you need to trust the chemistry, run the chemistry.
| Category | Pauling | Iambic Therapeutics |
|---|---|---|
| Core approach | Physics-based simulations (UniDock, GROMACS) | ML surrogates (NeuralPLexer, Enchant) |
| Molecular docking | UniDock (GPU-accelerated Vina) with explicit scoring | NeuralPLexer predicts bound poses via diffusion model |
| Structure prediction | P2Rank for pocket detection; uses experimental structures | NeuralPLexer -- joint protein-ligand structure prediction |
| MD simulation | Full GROMACS MD with checkpoint save/resume, ACPYPE parameterization | No molecular dynamics capability |
| Binding energy validation | MM-PBSA from explicit MD trajectories | Learned binding score -- no explicit free energy calculation |
| ADMET profiling | Integrated ADMET pipeline | Limited -- focus is on clinical endpoint prediction |
| Clinical endpoint prediction | Not in scope | Enchant predicts Phase I/II clinical outcomes |
| Validation methodology | PoseBusters pose QC, MolProbity structure checks | Benchmark against PoseBusters; ML confidence scores |
| Transparency of results | Full trajectory files, scoring logs, parameter tracking | Model predictions with confidence intervals |
| Compound screening scale | 10M+ compounds via auto-scaling Cloud Dataflow | Focused on lead optimization, not large-scale screening |
| File format handling | Automatic conversion: PDB, PDBQT, SDF, MOL2, CIF, SMILES | Proprietary input pipelines |
| Access model | Self-serve cloud platform, pay for compute | Internal pharma pipeline -- not publicly available |