Select any disease or upload patient genomic data. 57 AI agents across 5 tiers resolve targets, design multi-modal candidates, and deliver ranked therapeutics.
Every decision logged in immutable StageDecisionRecords. Every stage gate auditable. Every kill switch documented. Compliant with AIDD-GOV v0.1 open standard.
Select any gene from the left panel or click a gene chip to load analysis.
Powered by Claude · Anthropic API
CAMPAIGN HISTORY
DNA Composition
Active Trials
Select a gene to load clinical trials.
Drug Repurposing Check
Select a gene target to check for approved drugs.
Campaign configuration
Fragment budget (of 90)
45
RL iterations
4
Docking engine
AutoDock Vina 1.2.7 ✓
Gene interaction network
HistoryMarch 2026
Patient Intelligence
Select a gene target to view patient population estimates
Campaign Run
JB
1
Disease
2
Targets
3
Configure
4
AI Pipeline
Choose a disease
Or use the Genome Center for gene-first selection
Real-World Evidence
LIVE
Select a disease above to load real-world evidence
Select a disease to continue
Molecular targets
Resolved from DisGeNET, OMIM, MitoCarta 3.0
Gene
Protein
GDA score
Source
Registry
Run configuration
Fragments: 45
Iterations: 4
AI pipeline executing…
57 agents · DrugSynthAI governance engine
What just happened
57 AI agents scanned your genes, found binding pockets on their proteins, generated 163 drug candidates, and ranked them by therapeutic potential.
Each candidate is a small molecule that can enter a cell and bind a specific protein involved in your disease — altering its behavior to restore normal cell function. All candidates passed Lipinski drug-likeness rules, ADMET safety filters, and intellectual property novelty checks.
DEMO_001 · Rank 1
Blocks mitochondrial fragmentation
Binds DRP1 GTPase domain on the outer mitochondrial membrane, preventing fission. In disease neurons, this preserves ATP-producing mitochondria and reduces cell death.
DNM1LFission inhibitor
DEMO_002 · Rank 2
Activates cellular garbage collection
Activates PINK1 kinase, triggering Parkin-mediated mitophagy. Damaged mitochondria are tagged and destroyed before releasing toxic apoptotic signals.
PINK1Kinase activator
DEMO_003 · Rank 3
Prevents apoptosis trigger release
Stabilizes OPA1 on the inner membrane, tightening cristae junctions so cytochrome c cannot leak — the molecular switch that commits a cell to programmed death.
OPA1Fusion modulator
Results saved to platform registry
CMP-XXXXXXXX
SAVED
What happens next
V2 Scientific review
You review top 5 candidates vs. known clinical endpoints and literature.
The platform deploys 57 specialized AI agents organized into five tiers, each with a single responsibility and built-in quality gates. Pipeline agents (Tier 1) execute the S00 through S10 drug discovery stages in sequence. Validators (Tier 2) independently verify every stage output before advancement. Orchestrator services (Tier 3) manage scheduling, governance enforcement, and audit logging. Intelligence agents (Tier 4) operate on an independent cycle, scanning external databases for competitive intelligence. Personalized medicine agents (Tier 5) handle patient data ingestion, gene regulatory network inference, perturbation simulation, and multi-modal therapeutic generation including antibody design. 15 non-autonomous assistants are bound one-to-one with pipeline agents for delegated sub-tasks. No agent advances data to the next stage without a signed StageDecisionRecord.
Tier 1 — Pipeline Agents11 agents · S00–S10
Program Agent
S00
Initializes the campaign, validates all input registries, and establishes the governance context for the pipeline run.
Target Agent
S01
Identifies and validates molecular targets from clinical genetic databases. Resolves gene identifiers to protein structures.
Structure Agent
S02
Retrieves 3D protein structures and identifies binding pockets where drug molecules can attach to targets.
Generator Agent
S03
Designs novel drug candidate molecules using fragment-based assembly. Produces 163 unique candidates per campaign.
Expander Agent
S04
Expands the candidate library by generating structural analogs around lead scaffolds to maximize chemical diversity.
Docking Agent
S05
Runs physics-based molecular docking to compute binding free energies against validated receptor structures.
Safety Agent
S06
Predicts drug safety across absorption, distribution, metabolism, excretion, and toxicity. Applies kill conditions for unsafe candidates.
Novelty Agent
S07
Screens candidates against patent databases and approved drug libraries to ensure structural novelty and freedom to operate.
Scoring Agent
S08
Computes multi-objective composite scores balancing efficacy, safety, novelty, and manufacturability across all candidates.
RL Agent
S09
Runs iterative optimization to improve candidate scores against a multi-component reward function with convergence detection.
Packaging Agent
S10
Assembles the final campaign artifact bundle with ranked candidates, attestation records, and regulatory-ready documentation.
Tier 2 — Platform Validators15 agents
Independent quality control agents that verify the work of pipeline agents at every stage. Each validator checks specific output properties before the stage gate allows advancement. Validators cannot be overridden by the agents they audit.
ADMET Profiler
ADM_001 · Drug safety evaluation
Compound Curator
CCR_001 · Library integrity
Docking Analyst
DKA_001 · Binding validation
Fragment Strategist
FSA_001 · Fragment selection
IVVP Planner
IVL_001 · Validation protocol
IP/Patent Analyst
PLA_001 · Freedom to operate
Molecule Designer
MDA_001 · Candidate design
Peptide Designer
PDA_001 · Peptide mode
Integration Agent
PIA_001 · Pre-gate coordination
RL Architect
RLA_001 · Optimization design
Toxicity Analyst
TXA_001 · Structural alerts
Defect Modeler
DFM_001 · Disease mechanisms
Pathway Analyst
PAM_001 · Network mapping
Structural Biology
SBA_001 · Structure validation
Target ID Director
TID_001 · Evidence validation
Tier 3 — Orchestrator Services6 agents
Infrastructure agents that govern the pipeline: campaign lifecycle management, stage sequencing, governance header enforcement, compute scheduling, audit record persistence, and stage dispatch. These agents run continuously throughout every pipeline execution.
RunController
Central pipeline overseer
Campaign Manager
Lifecycle management
Policy Enforcer
Governance compliance
Resource Allocator
Compute scheduling
Audit Emitter
Immutable audit trail
Stage Executor
Stage dispatch
Tier 4 — Intelligence Agents2 agents
Independent cycle agents that scan external scientific databases, patent landscapes, and clinical registries. They generate research briefs that feed into the next campaign cycle. Not bound to the S00 through S10 pipeline sequence.
Research Intelligence Monitor
Literature and patent scanning
ROB-to-RLO Generator
Research order generation
Assistant Layer15 assistants
Non-autonomous helper agents bound one-to-one with pipeline agents. Each assistant handles delegated sub-tasks: input validation, output formatting, quality checks, and data packaging. Assistants cannot act independently and only execute when their parent agent delegates work to them. Their function is to keep pipeline agents focused on decision-making while assistants handle mechanical execution.
5/5 targets shared — programs are complementary, not redundant. Different compound scaffolds.
IP Sentinel
Loading USPTO...
Querying USPTO PatentsView API...
CYP Inhibition Risk
SwissADME
Click Run to predict CYP inhibition for top candidates.
BBB / CNS Penetration
Veber/Lipinski
Click Run to predict BBB penetration and CNS exposure.
Target Deconvolution
Off-target prediction for top candidates
Predicted off-target binding (in silico) — flags potential liabilities and unexpected polypharmacology
DEMO_001DNM1L GTPase inhibitor
DRP1 (primary)HIGH confidence
Dynamin-1 (GTPase family)LOW (Tanimoto 0.18)
Dynamin-2 (GTPase family)LOW (Tanimoto 0.21)
hERG (cardiac risk)CLEAR
CYP3A4 (metabolic)CLEAR
DEMO_002PINK1 kinase activator
PINK1 (primary)HIGH confidence
LRRK2 (kinome neighbor)LOW (kinome 0.19)
hERGCLEAR
CYP2D6FLAG — predicted interaction
Regulatory Audit Trail
21 CFR Part 11 ready
Complete audit trail for all platform decisions. Every agent action is logged with timestamps, input/output hashes, and a signed StageDecisionRecord. Export for regulatory submission or internal QA review.
Total SDRs issued18 (9 per campaign)
Compliance violations0
Audit log entries163 agent actions
Electronic signatureSHA-256 hash chain
TraceabilityGene → candidate → SDR
Competitive Landscape
Idle
DNM1L moat
—
PINK1 moat
—
Big Pharma signal
—
Market gap
—
Click Refresh above to load competitive landscape data from OpenFDA and ClinicalTrials.gov.
Program value estimation
Indicative · not financial advice
$2–8M
Target-ID + Hit-Gen value
Industry standard for computational campaign delivering 163 ADMET-qualified hits
$15–40M
Lead series value at IND
If DEMO_001 advances through preclinical with confirmed MOA
$100–400M
Phase II POC milestone
Typical partnering deal for validated CNS/mitochondrial target in rare disease
FDA FAERS signals for pathway-adjacent approved drugs · Amber >100 reports · Red = hepatotoxicity flag
Click Refresh to load FAERS adverse event data.
Active Clinical Trials — Target Intelligence
ClinicalTrials.gov API v2 · Green = first-mover · Amber = active competition
Click Refresh to query ClinicalTrials.gov for active programs targeting your gene set.
Analytics
CMP-40BB5083 · CMP-76D2FBEC · ChEMBL 33 live
JB
Best composite
0.7654
CMP-76D2FBEC · PINK1
Hit rate
76.1%
124/163 Tier A ADMET
RL improvement
+7.2%
0.76 → 0.82 · 4 iters
Best docking
−8.0 kcal
DNM1L GTPase · DEMO_001
IP novelty
100%
All Tanimoto < 0.35
Compliance
9/9
SDRs PASS · 0 violations
Optimization Performance
How well did the AI optimize drug candidates? The RL convergence curve shows whether the reward function found a stable optimum. The score distribution reveals if optimization enriched for high quality candidates or left a flat, uniform spread.
■ Top quartile (>0.72)■ Mid■ LowerMean: 0.694 · SD: 0.031
Safety and Binding
Can these molecules survive the body and bind their targets? ADMET tiers classify drug safety. Docking energy measures binding strength. Molecular weight must stay within drug-like ranges (Lipinski Rule of Five).
ADMET Profile
Tier A (all criteria optimal)124 (76.1%)
Tier B (1 boundary criterion)39 (23.9%)
Tier C (failed)0 (0%)
PAINS alerts: 0 of 72 filters triggered
Docking ΔG Distribution
Best affinity−8.0 kcal/mol
Mean affinity−7.3 kcal/mol
Cutoff threshold−6.5 kcal/mol
Pass rate163/163
MW Distribution
Mean MW308.4 Da
Range246–398 Da
Lipinski ≤500 Da163/163
Fragment-like ≤300 Da48 (29.4%)
Campaign Intelligence
Side by side comparison of active campaigns. Which protein targets are most druggable? Higher druggability scores indicate better binding pocket geometry and historical success rates for that target family.
Drug-likeness compliance, intellectual property novelty, and wet lab readiness. All candidates must pass Lipinski/Veber criteria. Tanimoto scores below 0.35 confirm structural novelty versus every approved drug in ChEMBL 33.
Lipinski compliance breakdown
MW ≤500 Da
163/163
LogP ≤5
163/163
HBD ≤5
163/163
HBA ≤10
163/163
TPSA ≤140Ų
163/163
RotBonds ≤10
163/163
All candidates satisfy full Lipinski + Veber criteria
DEMO_001, DEMO_002 — patch clamp recommended pre-in vivo
QueuedMicrosomal stability
HLM half-life prediction: all candidates >30 min (estimated)
Mitochondrial Pathway Network — SBGN-style
31 edges · 25 nodes
Drag nodes · Scroll zoom · Click to inspect
OXPHOSDynamicsMetabolismCentral dogma
Advanced Genomic Visualizations
High dimensional views of the candidate chemical space. The 3D surface maps score landscapes across targets. PCA and tSNE projections reveal clustering patterns. The volcano plot highlights candidates with both high selectivity and high composite scores.
3D Gene Expression Landscape
X: Target index Y: Pathway cluster Z: Selected metric Drag to rotate
PCA — Candidate Feature Space
PC1/PC2 · ADMET tier
Cluster Map — Chemical Space
tSNE-style · size=score
Volcano Plot — Selectivity vs Composite Score
size=novelty · color=IP risk
Validation and Synthesis
Scientific validation status, binding stability predictions, synthetic feasibility analysis, and network perturbation effects. These panels evaluate whether computationally designed candidates are scientifically viable for wet lab synthesis and testing.
V2 Scientific Validation
Tanimoto threshold: 0.35 (novel)
Loading V2 validation data...
Binding Stability Ranking
Geometry proxy · GPU MD via Colab
Click "Load Stability Ranking" to compare static docking rank vs estimated binding residence time.
Retrosynthesis Overview
BRICS proxy · ASKCOS/IBM RXN for exact routes
Click "Load Retrosynthesis" to view synthetic accessibility tier for each IVVP Tier 1 candidate.
Network Analysis — Mitochondrial Signaling Graph
8 nodes · 10 edges · NPD + cascade
Click "Load Network" to visualize how each candidate perturbs the mitochondrial signaling network.
Total weight: 1.00 · 5 kill conditions · RWD_007 active Sprint 37 onwards · CMP-AGING-001 first 7-component campaign
Governance Analytics
Constraint Sensitivity
Click Load to run sensitivity analysis.
Cross-Campaign RL Convergence
Click Load to overlay all campaign curves.
Campaign Governance Diff
vs
Select campaigns and click Compare.
DECISION INSIGHTS
Select a campaign to see insights.
GOV-FM Radar
AI Research Chat
Powered by Claude · Context-aware drug discovery assistant
JB
Conversations
Today
Campaign CMP-40BB5083
Mitochondrial · 8 msgs
DrugSynthAI Research Assistant
CMP-40BB5083 · Context loaded
Hello Dr. Borges. I have full context on campaign CMP-40BB5083 — 163 candidates, top composite score 0.82 for DEMO_001 targeting DNM1L. What would you like to explore?
These are the governed knowledge artifacts that drive every campaign. Click any registry to inspect its contents, view entries, or download the raw YAML.
Platform Data Registries
Registries are the structured data files that govern every decision in the drug discovery pipeline. Each registry is a YAML file containing validated records: gene targets, molecular fragments, compound candidates, scoring weights, and compliance rules. The AI agents read from these registries at each pipeline stage. Editing a registry changes what the platform optimizes for.
22 of 25 targets are fully docking-ready with validated receptor files. 3 targets pending PDBQT preparation (AlphaFold structures available, pocket detection complete). All governance registries locked at v1.
DATA GAPS
3 pockets need PDBQT conversion (SLC25A4, TFAM, POLG2). Candidates from these targets will skip S05 docking until receptor files are prepared. No impact on the remaining 22 active docking channels.
RECOMMENDATIONS
1. Prepare PDBQT for 3 pending pockets to reach 100% coverage. 2. Review constraint_policy kill conditions before next campaign. 3. Consider expanding fragment library for peptide mode targets.