Backend truth + dated audit
The detailed registry is runtime-backed; the six-capability maturity map is explicitly dated so snapshot and live state are never confused.
BAYESPHARMA INNOVATOR
Connecting to live registry
Scientific Capability Registry
A live, evidence-aware view of what is actually inside BayesPharma Innovator—including BayesARC's auditable Protein Engineering sequence-design and binder-candidate triage control plane and its engine-readiness states—plus a separate BayesPeaQ QMS capability view and a technology radar showing which alternatives could make each workflow stronger.
Runtime tells you what is available now.
Upgrade radar shows what should be benchmarked next—not what should be trusted automatically.
Availability ≠ qualification
Installed, configured and scientifically qualified remain separate concepts—including every Protein Engineering engine.
Benchmark before promotion
A newer model becomes preferred only after it wins in the intended context of use.
No secrets exposed
Credentials, private worker URLs and connection strings never belong in this registry.
HEADLINE CAPABILITY MATURITY · AUDITED 29 AUG 2026
What we genuinely have—and what still needs integration.
This product-level audit answers the six capability questions customers are most likely to ask. It deliberately distinguishes production execution from governed control planes, partial evidence workflows, deployment-gated engines and scientific qualification.
01 · STRUCTUREOperational integration
Protein structure prediction
Now: sequence-to-structure execution through the ESMFold-compatible ESM Atlas API, with PDB output, residue count, mean pLDDT/confidence, NGL visualization and direct docking handoff. Current sequence limit is 400 residues and external processing requires explicit consent.
Integration path
Keep ESMFold as the current bounded route while the Protein Engineering registry separately tracks ESMFold2, Chai-1, Boltz-2 and other challengers. Persist provider/version, confidence, inputs and provenance, and benchmark every engine before promotion.
02 · DOCKINGProduction operational
Molecular docking
Now: real AutoDock Vina with Meeko/RDKit support, single and batch campaigns, pose outputs, target qualification, receptor-preparation gates and the Discovery Campaign funnel. Scores remain computational ranking evidence, not measured binding affinity.
Integration path
Extend the strongest current capability with governed ensemble docking, alternative rescoring, metal/covalent-aware routes where appropriate, and MD/MM-GBSA qualification—without treating any one score as experimental proof.
03 · CHEMISTRYOperational core · deep engines deployment-gated
Generative chemistry
Operational now: real RDKit/BRICS generation, canonical-SMILES deduplication, medicinal-chemistry/property constraints, scaffold diversity, asynchronous million-scale campaign orchestration, persistence/pagination/cancellation and Discovery Campaign/Vina handoff. Already integrated in code: self-hosted GenMol, Pocket2Mol and DiffSBDD execution adapters plus provider-neutral REINVENT/pocket-generator routes. Deep engines are only promoted to available when their model checkout, checkpoint, compatible Python runtime and GPU are actually present.
Next qualification gate
Deploy and benchmark the self-hosted GPU/checkpoint stack, persist exact model/checkpoint provenance, add reproducible target-conditioned benchmarks, and evolve the existing multi-engine architecture toward synthesis-aware multi-parameter optimization and active-learning feedback. Do not label a deep generator production-operational until runtime readiness and scientific qualification both pass.
04 · GENOMICSPartial
Genomic analysis
Now: live variant annotation and genetic target-evidence retrieval through Ensembl VEP, Open Targets, gnomAD, ClinVar/NCBI and GWAS Catalog. This is scientifically useful evidence analysis, not a full WGS/WES, RNA-seq or multi-omics pipeline.
Integration path
Create a governed genomics service for VCF ingestion and sample/cohort workflows, then add RNA-seq QC plus DESeq2/limma execution. Feed qualified genetic evidence into Target Readiness so genomics becomes part of the programme golden thread.
05 · BIOLOGICSGoverned platform operational · ProteinMPNN deployment-gated
Protein engineering & biologics design
Operational now: an auditable protein sequence-design and binder-candidate triage control plane with tenant-scoped records, strict mode-specific input contracts, persistent campaigns and designs, deterministic sequence QC and residue-level liability analysis, unit-aware/applicability-aware Pareto ranking, authenticated append-only reviews, immutable Design Passport versions, provenance-rich artifacts and metrics, Design Explorer, and NGL structure inspection when a real structure is available. Merged execution architecture: tenant-scoped Postgres persistence, a durable Render Key Value job queue, retry/job-attempt and idempotency controls, and an exact-commit ProteinMPNN adapter/worker path with model, weights, input, seed, parameter and artifact provenance. Current execution boundary: ProteinMPNN is merged and pinned but is not represented as production-operational until its dedicated worker deployment and engine-qualification gate are green.
First release + next qualification gate
Target structure + chains + epitope → input validation → ProteinMPNN sequence design → sequence/developability QC → structure/complex validation → Pareto ranking → authenticated human review → immutable Passport. Green the pinned ProteinMPNN worker first, then verify same-input/model/seed reproducibility and publish an engine qualification record. Add bounded MHC-I presentation-risk screening and antibody/nanobody/TCR structure workflows as independently qualified modules. Keep RFdiffusion de novo binder-backbone generation and Boltz/Chai independent complex validation behind GPU and context-of-use qualification gates. BayesARC does not claim production de novo binder generation until GPU execution and independent structural validation pass.
06 · TRANSLATIONNot implemented
Biomarker discovery
Now: Biomarker & Endpoint Studio evaluates supplied biomarkers, assay/readiness considerations, BEST categories, endpoints and regulatory evidence. It does not yet discover new biomarkers from patient or omics data.
Integration path
Build a separate discovery pipeline for expression matrices, case/control analysis, feature selection, nested validation, ROC/calibration, survival modelling, pathway enrichment and external-cohort validation. Keep discovery and evidence-readiness assessment as distinct stages.
Why the status can differ from the live registry: this section is a dated end-to-end capability audit; the registry below reports current deployed components, engine configuration and qualification state. Generative Chemistry has an operational core while deeper generators remain deployment/qualification-gated. Protein Engineering now has the governed data, job and evidence architecture on main—including tenant isolation, Postgres/Key Value persistence, strict schemas, provenance, append-only authenticated review and immutable Passports. ProteinMPNN is the first real structure-conditioned sequence-design execution path, but remains deployment/qualification-gated until its dedicated worker is green and benchmarked. RFdiffusion, Boltz and Chai remain GPU/independent-validation gates; no production de novo binder-generation claim is made.
BAYESPEAQ · PHARMACEUTICAL QMS CAPABILITIES
Quality work that stays inspectable from event to decision.
BayesPeaQ is the standalone pharmaceutical QMS within Bayes Pharma GenRx. This section is intentionally separate from the Innovator scientific registry: it describes governed quality-system capabilities, evidence controls and validation tooling rather than drug-discovery model capability.
01 · CORE QMSImplemented
End-to-end governed quality workflows
Now: controlled workflows for deviations, OOS/OOT, investigations/RCA, CAPA, change control, documents, training, QRM, supplier quality, audits, validation, complaints, APQR/PQR, batch disposition, recalls, data-integrity investigations and related quality domains.
Governance
Validated state vocabularies, controlled transitions, role requirements, configurable gates and SLA metadata prevent workflow skipping while keeping human QA authority explicit.
02 · DATA INTEGRITYImplemented
Electronic records, signatures & auditability
Now: exact-content hashing, append-only audit history, one-time reauthentication, signature meaning/rationale, segregation of duties, distinct-signature rules and governed approval/release transitions.
Boundary
These are technical controls designed to support regulated use. They are not a claim of formal customer validation, Part 11 certification or Annex 11 certification.
03 · ACCESSImplemented
Role-based quality authority
Now: QA, QC, Investigator, CAPA Owner, Production, Engineering, Warehouse/Supply Chain, Regulatory, Validation, Training, Site Head, Authorized Person, Auditor and controlled administration personas with least-privilege boundaries.
Separation of duties
Auditor and external-inspector views remain read-only; administrative access does not automatically grant controlled-record approval or release authority.
04 · EVIDENCEEvidence-native
Evidence completeness & inspection readiness
Now: controlled evidence objects, linked quality records, audit-trail review, validation traceability, Supplier 360, document-to-training linkage and deterministic investigation-evidence completeness scoring.
Inspector view
The system is designed so a reviewer can follow one event through evidence, investigation, decisions, signatures, linked CAPA/change/training records and history without relying on an AI-generated narrative alone.
05 · QUALITY AIHuman-governed
Quality Copilot & role-agent validation
Now: controlled-record retrieval, recurrence signals, evidence-gated RCA support, CAPA/change-impact support, inspection-readiness assistance and governed multi-model role-agent validation using private Qwen plus configured external models where permitted.
Non-negotiable rule
AI recommends and challenges. BayesPeaQ preserves evidence and workflow controls. Authorized people sign, approve, close and release.
06 · VALIDATIONFramework implemented
Golden Workflow + QMS benchmark framework
Now: an equipment-failure Golden Workflow assessor, 10 representative GMP scenarios and a 100-case synthetic Golden Dataset with known reference labels for risk, RCA, CAPA, evidence, recurrence and regulatory-assessment expectations.
Measured outputs
The framework can score risk accuracy, RCA reference recall, CAPA coverage, recurrence precision/recall, unsupported claims and correct abstention from final GMP decisions. Actual model-performance claims require completed benchmark runs.
Product boundary: BayesPeaQ belongs to Bayes Pharma GenRx, not the Innovator scientific execution stack. It is shown here so the company capability view clearly separates scientific capabilities from quality-system capabilities. Open the dedicated BayesPeaQ product page ↗
CURRENT PLATFORM STATE
What is actually inside Innovator
Search the complete public workspace catalog and the runtime-backed scientific technology layer. Protein Engineering engines expose runtime readiness and scientific qualification separately.
Reading Innovator backend state
Checking registry declarations, product workspaces and configured scientific runtimes.
Live registry is temporarily unavailable
The page will not substitute a marketing list when backend truth cannot be verified.
No capabilities match these filters.
Try a broader search or clear one of the status/category filters.
| Capability | Actual tool / model / system | Runtime status | Scientific status | What it does | Details |
|---|
BEST-IN-CLASS TECHNOLOGY RADAR
How we make Innovator better—not just bigger.
For high-value capabilities, the backend keeps a governed challenger set: alternatives that can be better for a specific scientific or operational job. The rule is simple: keep the current production tool until the challenger proves a material gain on BayesPharma benchmarks.
Loading best-in-class challengers…
READ THE STATUS CORRECTLY
Software presence is not scientific evidence.
A model can be loaded, a worker can be healthy, and a database can be reachable while a specific target, endpoint or context still requires qualification. The registry keeps these states deliberately separate.
Live / integratedBackend/runtime evidence says the capability is available.
ConfiguredThe integration is configured; this does not prove current scientific suitability.
Qualification requiredA target-, endpoint- or context-specific scientific protocol is still required.
Unavailable / degradedThe current backend state does not support representing it as live.
SCIENTIFIC TRANSPARENCY
Inspect the capability—and the challenger—before trusting the output.
BayesPharma Innovator is designed to preserve what generated a claim, what evidence class it belongs to, which alternative could improve it, and where human scientific judgment is still required.