AlphaDock OS is the structure and affinity workbench inside Bayes Pharma.ai Discovery. It connects target context, structure hierarchy, AlphaFold-guided docking, AutoDock Vina pose output, ADMET triage, AI explanation, verifier notes, and reviewable decision dossiers.
AlphaFold + Vina evidence
Verifier and provenance layer
AI explanation
ADMET triage
PK/PD handoff
AlphaDock sits in the target-to-lead sequence as the structure, pose, ADMET, and confidence layer feeding the same decision record.
Open workflows can use open parameterization paths, while commercial force fields are positioned as licensed integration paths when applicable.
Open ligand parameterization paths for prepared small-molecule simulation and review workflows.
Open LigandsProtein, peptide, membrane, and biomolecular simulation options when system preparation is controlled.
BiomoleculesPosition as a licensed integration path where the customer has the appropriate rights and environment.
Licensed PathAlphaDock preserves what was run, why it was run, what the model assumed, where methods disagree, and what next experimental question follows.
Affinity estimate, interpretation, pose rank, method spread, and source method are shown together instead of as an isolated score.
3D inspection, pose source, docking box assumptions, and receptor-preparation caveats stay visible for reviewer challenge.
Lipinski, hERG, property risk, and developability indicators help decide whether docking evidence is actionable.
Compare mode exposes consensus and disagreement between AlphaFold-guided, enhanced ML, and Vina evidence.
Hypothesis-generating boundaries, pocket selection, cofactors, protonation, water handling, and validation needs remain explicit.
Send prioritized compounds into synthesis feasibility, assay planning, lead optimization, or PK/PD translation.
Docking and ML estimates are useful for prioritization, but decision-grade use requires structure preparation, reviewer challenge, and experimental validation.
AlphaFold-guided prediction, enhanced ML scoring, real AutoDock Vina pose output, ADMET triage, AI explanation, and NGL pose inspection.
Pocket selection, protonation state, cofactors, metals, structural waters, docking box dimensions, assay context, and analog-series relevance.
Outputs are hypothesis-generating and do not replace medicinal chemistry review, assay confirmation, MD/FEP validation where needed, or development-team judgment.
Use Vina for real pose output. Compare mode can blend AlphaFold, Enhanced ML, and Vina evidence.
Target-aware prediction using existing AlphaFold endpoint.
Affinity estimate, interpretation, pose files, and diagnostics.
Select a mode, review target and ligand inputs, then run the prediction.
Evaluate Lipinski properties, hERG risk, and overall developability risk for the current ligand.
RDKit synthetic-accessibility score, functional-group disconnection hints, and a purchasable-precursor similarity check for the current ligand.
Live tractability, disease association, safety liability, and known-drug evidence from the public Open Targets Platform API.
Every completed docking, ADMET, and synthesis-triage run in this session is appended to a hash-chained, exportable evidence ledger.
Loads the real AutoDock Vina pose when available; otherwise falls back to sample pose.
Use foundation models and biomedical agents as structured workers inside the Discovery record, with confidence, disagreement, and applicability shown beside the result.
Target-disease evidence, tractability, safety, known drugs, and disease biology
Live · Evidence GraphComplex structure and affinity signal for rapid triage, not a final FEP replacement
Structure/AffinityRetrosynthesis routes, purchasable precursors, policy control, and route confidence
SynthesisTherapeutic-property specialists for toxicity, BBB, reaction, and explanation tasks
EnsembleOptional pose and scoring workers for method agreement and pose-quality challenge
Docking EnsembleGPU-scale biomolecular model infrastructure when fine-tuning or distributed inference is justified
InfrastructureThe priority is not more isolated models. It is target context, structure consensus, synthesis feasibility, provenance, and validated learning loops.
Live grid-based pocket detection (LIGSITE-style) auto-fills the Vina docking box; AlphaFold/Enhanced ML/Vina consensus is available in Compare mode.
LiveEvery run is appended to a hash-chained provenance dossier you can inspect and export as JSON.
LiveLive RDKit synthetic-accessibility score, disconnection hints, and route-risk rating. Full retrosynthesis search (AiZynthFinder) remains roadmap.
LiveRun OpenMM or GROMACS only after protein prep, protonation review, ligand parameters, and QC gates.
NextUse relative free energy on prepared congeneric series with uncertainty and cycle-closure checks.
NextClose the loop with verified assay results, temporal validation, leakage checks, and acquisition strategy.
NextFlexible docking for peptide therapeutics when structure preparation and sampling limits are explicit.
LaterAdopt GPU-scale biomolecular infrastructure after workloads justify fine-tuning or distributed inference.
LaterThe differentiator is traceability: every score carries source context, assumptions, uncertainty, reviewer cautions, and downstream action.
| Decision need | Point-tool pattern | AlphaDock evidence workflow |
|---|---|---|
| Structure choice | One receptor model is often used without enough context. | Experimental PDB, AlphaFold, and roadmap complex-prediction sources are separated and labeled. |
| Pose evidence | Docking score is exported as a standalone result. | Affinity, pose rank, method, docking box, and viewer state remain in one dossier. |
| Method agreement | Different scoring methods are compared manually. | Compare mode can expose AlphaFold-guided, enhanced ML, and Vina agreement or spread. |
| Developability context | ADMET is checked in a separate workflow. | ADMET and drug-likeness flags sit beside docking evidence before prioritization. |
| Scientific challenge | Assumptions may live in notes or slides. | Pocket, protonation, cofactors, waters, and experimental validation needs stay visible. |
| Handoff | Results move by screenshots or copied tables. | Evidence can feed synthesis feasibility, lead optimization, assay planning, and PK/PD translation. |
Prediction feedback enters a versioned learning loop after assay results are verified, leakage checks are complete, and the acquisition strategy is reviewable.
Vina mode uses the latest AlphaFold PDB URL, prepares receptor and ligand files, runs AutoDock Vina, returns ranked poses, and exposes the real generated pose to NGL.
The default auto-centered box is useful for a technical run but not enough for production biology. Decision-grade use requires pocket selection, protonation control, cofactors/metals, waters, and experimental validation.
Start with a target and ligand, inspect binding evidence, capture ADMET flags, and route the package into the next Discovery decision.
Programme context stays with you across Innovator · Evidence classes remain explicit · Heavy workflows require confirmation · Do not share identifiable patient data.