Patent Pending US 64/018,624

Governed Translational
Decision Infrastructure

69 autonomous agents. Calibrated pipeline. Multi-backend docking (AutoDock Vina + Boltz-2). De novo molecule generation (BRICS + REINVENT4). From disease target to safety-screened, novel therapeutic candidates with full governance audit trail.

Launch Platform Read Documentation
69
AI Agents (5 Tiers)
2,032
Tests Passing
41
Novel Molecules Designed
25
Docking Targets (Mito)
17
Published DOIs
Governed Pipeline

S00 to S10. Every stage gated.

Each stage requires a StageDecisionRecord before advancement. No candidate advances without passing every kill condition. The governance trail is immutable.

DISEASEYAML config
→
S00Init
→
S01Targets
→
S02Structure
→
S03Fragments
→
S04Analogs
→
S05Docking
→
S06ADMET
→
S07Toxicity
→
S08Novelty
→
S09Scoring
→
S10RL + IVVP
→
CRODispatch
→
MDDynamics
→
DELDelivery
Five Service Lines

One platform. Five translational services.

The same governed infrastructure serves distinct use cases depending on what enters the pipeline and who receives the output.

Computational Audit

TCDR-as-a-Service

Submit a compound. Receive a governance-grade Therapeutic Candidate Decision Record: docking profile across 25 targets, ADMET, 7-dimension MCDA scoring, 3 weight scenarios, comparative landscape position. Calibration evidence accompanies every delivery. For pharma BD teams, biotech pipeline evaluation, and foundation diligence.

De Novo Generation

Campaign-as-a-Service

Specify a target, constraints, and seed compounds. The platform executes S03 through S10 and delivers 3 to 8 novel candidates with complete TCDR records, docking scores, ADMET profiles, and novelty assessment. Novel molecules are the product. For pharma pipeline expansion and biotech second-generation candidates.

Precision Medicine

Variant-Level Scoring

Submit a patient genotype. Tier 5 agents map the variant to affected protein subunits, identify druggable targets, score existing therapeutics against the variant context, and return ranked recommendations. For clinicians, genomics programs, and rare disease foundations.

Landscape Intelligence

Competitive Analysis

Submit a therapeutic area. The platform evaluates every relevant compound, ranks by MCDA composite, identifies translational gaps, and produces a strategic intelligence report. For investors, venture philanthropy, and pharma BD portfolio evaluation.

Gene Therapy Evaluation

TCDR-MI (v2.0, 12 Field Groups)

The same MCDA framework applied to gene therapy candidates: AAV vector identity, tropism, payload, transduction efficiency, immunogenicity, and manufacturing scalability. Modality-agnostic ranking across small molecules, peptides, antibodies, and gene therapy in a single framework.

Computational Backends

Multi-backend architecture. Calibrated before every campaign.

Pluggable backends at S03 (generation) and S05 (docking). Ensemble mode cross-validates predictions across independent methods. Calibration protocol runs a known reference compound and verifies 34+ checks before any campaign fires.

S03 — Molecule Generation
BRICS — Fragment recombination from seed compounds. Deterministic, reproducible, retrosynthetically accessible. Same seed produces same output every time.

REINVENT4 — Neural generative model (AstraZeneca, Apache 2.0). RNN pretrained on ChEMBL. Three-phase: transfer learning on seeds, staged reinforcement learning with campaign constraints as scoring, de novo sampling. Explores chemical space beyond seed fragments.

Hybrid — Both backends pooled, deduplicated by InChIKey. Maximum chemical space coverage.
S05 — Binding Affinity Prediction
AutoDock Vina 1.2.7 — Physics-based docking. Requires receptor PDBQT files. 6 of 22 mitochondrial targets produce real scores; 12 fail upstream PDBQT conversion.

Boltz-2 — Diffusion model (MIT CSAIL + Recursion, MIT license). Takes protein sequence + ligand SMILES directly. No PDBQT needed. Approaches FEP accuracy at 1000x lower compute cost. Eliminates the 55% PDBQT failure rate. Predicts affinity + confidence (ipTM, pLDDT).

Ensemble — Both backends. Divergence above 1.5 kcal/mol triggers kill condition. Cross-validation on shared targets.
Nine Computational Layers

167 endpoints. Every one real.

Every layer produces real computation, real data, and real audit records.

Governance · 9 endpoints

StageDecisionRecords, audit trails, kill switches, AIDD-GOV compliance, GOV-FM self-improving scorer. Every decision immutable.

Science · 6 endpoints

AutoDock Vina 1.2.7 real docking, RDKit ADMET, BRICS retrosynthesis, 3D conformers, 72-rule toxicity SMARTS, patent search.

Evidence · 1 endpoint

Orphanet, GARD, ClinVar, HPO queried in parallel. Population-level epidemiological data grounding every campaign.

Synthesis · Phase A · 4 endpoints

ASKCOS retrosynthesis with BRICS fallback, commercial availability, CRO brief generator, assay protocol library.

Patient Intel · Phase B · 5 endpoints

Genotype-to-compound matching, population estimation (gnomAD + Orphanet), companion diagnostics, foundation data.

Molecular Dynamics · Phase C · 7 endpoints

OpenMM MD simulation, FEP binding affinity (ΔΔG), system preparation, GPU detection, trajectory analysis (RMSD/RMSF).

Delivery · Phase D · 6 endpoints

TPP+ conjugate design (850x matrix accumulation), 7-strategy prodrug designer, 5-route formulation spec, CMC parameters.

Regulatory · Phase E · 6 endpoints

Pre-IND meeting packages, CMC dossier sections, nonclinical summaries, full IND assembly (CTD Modules 1-5), FDA readiness scoring.

External Intelligence · 4+ APIs

OpenTargets GraphQL, ChEMBL, FDA FAERS, ClinicalTrials.gov, AlphaFold, UniProt, STRING DB, NCBI Entrez. Live data feeds.

Patient-Specific Generative Therapeutics

From patient genomic profile to ranked therapeutic candidates.

Upload tumor genomics, RNA sequencing, or clinical lab reports. The platform infers the patient's disease-specific regulatory circuitry, simulates interventions, and generates multi-modal therapeutic candidates, all under governed audit.

1

Upload

VCF files, FoundationOne reports, Tempus results, RNA-seq expression matrices. Parsed into a unified patient profile automatically.

2

Analyze

Patient-specific gene regulatory network inference. Perturbation simulation identifies highest-impact therapeutic targets with lowest off-target risk.

3

Generate

Multi-modal candidates: small molecules, peptides, and antibodies. Ranked by confidence, screened for safety, cleared for IP. Full patient therapeutic report.

Access

Currently in closed beta. Request early access below.

BETA TEST MODEPricing will be announced upon general availability. All beta testers receive priority access and founding member benefits.
Explorer
Explorer

Academic researchers, thesis work, non-commercial use. Full governed pipeline. Apply for beta access.

NEW
Precision Medicine
Clinical

Clinicians, genomics programs, oncology centers. Patient data upload, regulatory graph inference, multi-modal candidates.

Professional
Professional

Biotech startups, drug discovery labs, CROs. Unlimited campaigns, antibody design, API access, CRO dispatch.

Enterprise
Custom

Pharma, large biotechs, hospital networks. Private deployment, multi-user, SLA, custom disease domains, white label.

All tiers include full governance audit trail. Patent Pending. Contact jyborges@bu.edu for Enterprise pricing.

Research Output

10 manuscripts. 17 DOIs. 1 patent. 1 open standard.

Every manuscript passed R1 revision, R2 independent review, and IP audit. Preprint versions redacted per patent security protocol.

SSRN

10 Individual DOIs

M01-M10 on SSRN. Pathway mapping, compound design, docking, ADMET, RL optimization, platform architecture.

Preprints

Preprint Deposits

5 ChemRxiv LIVE. Research Square DOI 10.21203/rs.3.rs-9305508/v1. SSRN 10/10. arXiv pending endorsement.

AIDD-GOV

Open Governance Standard

Apache 2.0. 10 schemas, 3 conformance levels. github.com/fxmedus/aidd-gov

USPTO

US Provisional 64/018,624

Filed March 27, 2026. Two amendments, 51 pages. Covers 69-agent 5-tier architecture, Phases A-E, SIRLP, AIDD-GOV, patient-specific generative therapeutics.

First Domain Application

Mitochondrial Therapeutics: From Assessment to Design

25 gene targets. 201 compound library. 9-compound comparator landscape scored. 41 novel molecules designed by the engine. Calibrated pipeline with governance audit trail.

"Every existing Complex I rescue compound has a critical translational flaw: PAINS alerts, excessive lipophilicity, injection-only delivery, or insufficient BBB penetration. The platform encoded all six failure modes as hard constraints and generated 500 candidates. 20 survived. All 20 pass every constraint that eliminates every comparator. A new chemotype for mitochondrial disease."

CAMPAIGN-CI-RESCUE-BBB-v1 — First generative campaign output, May 2026
Precision Medicine

Patient-Specific Therapeutics

Upload patient molecular data. The platform infers disease regulatory circuitry, simulates interventions, and generates optimised candidates across small molecules, peptides, and antibodies.

Patient Data Ingestion
VCF, FoundationOne, Tempus, and Guardant reports parsed and unified into a standardised PatientProfile. Disease inference and gene panel selection are fully automated.
Regulatory Graph
Patient-specific gene regulatory networks inferred by GRN_001. Master regulators, vulnerability nodes, and enhancer–promoter interactions visualised in the interactive graph viewer.
Multi-Modal Design
Small molecules, stapled peptides, and therapeutic antibodies ranked on the same unified reward function. Patient-specific candidates across all three modalities in a single campaign.
Collaborate

Built for clinicians, foundations, CROs, and researchers

Patient-specific therapeutic reports. CRO-ready dispatch packages. Governance audit trails. Foundation proposals with real patient numbers. If you run a precision oncology program, genomics center, or rare disease foundation, we want to hear from you.

jyborges@bu.edu Read the Manual