Pauling runs computational chemistry at scale -- 10M+ compound screens on auto-scaling Google Cloud Dataflow, not one molecule at a time. Validated physics-based pipelines (UniDock, GROMACS, P2Rank, MolProbity, PoseBusters) with full quality control and conversational access. Rowan is an AI-native platform built for interactive, single-molecule calculations -- convenient for exploring one structure at a time, not for screening libraries at scale.
| Category | Pauling | Rowan |
|---|---|---|
| Computational approach | Physics-based: UniDock (Vina), GROMACS, MM-PBSA -- peer-reviewed methods with known accuracy profiles | AI/ML-first: surrogate models and learned potentials for speed |
| Docking | UniDock (GPU-accelerated Vina) -- validated scoring, scales to 10M+ compounds | AI-accelerated docking with learned scoring functions -- single-molecule jobs, not library-scale screens |
| Molecular dynamics | Full GROMACS MD: equilibration, production, trajectory analysis, MM-PBSA | Limited or surrogate-based MD -- not classical all-atom simulation |
| ADMET profiling | Integrated ADMET predictions within the pipeline | ADMET predictions available |
| Quality control | MolProbity for structure validation, PoseBusters for docking pose QC -- every result is auditable | Limited validation tooling for QC of generated poses |
| Pocket detection | P2Rank integrated -- automatic binding site identification from structure | Binding site prediction available |
| Validation methodology | Every pipeline step uses published, peer-reviewed tools with known error bounds | ML model accuracy depends on training data coverage -- harder to audit edge cases |
| Scaling | Google Cloud Dataflow with Apache Beam -- auto-scales to hundreds of workers for 10M+ compound screens | Cloud-based, but oriented to interactive one-at-a-time calculations |
| Cloud infrastructure | Google Cloud with managed storage, job tracking, checkpoint/resume | Cloud-native platform |
| Interface | Conversational AI -- describe experiments in natural language | Modern web interface with AI assistance |
| Private data | Owner-scoped molecule catalogs (owner_id) with full data isolation | Platform-level data management |
| Cost | Pay-per-use compute | Subscription-based pricing |