BioBrain — MindNetQ

Why BioBrain Matters

Cortex decides the scientific intent. BioBrain turns that intent into biomolecular evidence and molecule candidates.

Without BioBrain, Cortex relies on external scoring functions it doesn't understand or control. With BioBrain, Cortex owns its understanding of targets, pockets, binding, safety, synthesis, and generation — and improves each one from every pipeline outcome.

The moat: BioBrain starts from public data (PDBbind, ChEMBL, USPTO, ToxCast) — same as everyone. But it continuously retrains from our failure attributions and pipeline outcomes. After 6 months, our binding model knows things no public model can.

┌─────────────────────────────────────────────────────────────┐ │ CORTEX │ │ hypothesis + target + constraints │ └───────────────────────────┬─────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ BIOMOLECULAR BRAIN ORCHESTRATOR │ │ │ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │ │TargetNet │ │ PocketNet│ │ MolGen │ │ │ │ 70% ✓ │ │ 40% │ │ 55% ✓ │ │ │ └────┬─────┘ └────┬─────┘ └────┬─────┘ │ │ │ │ │ │ │ ┌────┴─────┐ ┌────┴─────┐ ┌────┴─────┐ │ │ │ BindNet │ │SafetyNet │ │ SynthNet │ │ │ │ 60% ✓ │ │ 75% ✓ │ │ 20% │ │ │ └──────────┘ └──────────┘ └──────────┘ │ │ │ │ Returns: BiomolecularOpinion + CandidatePackages │ └───────────────────────────┬─────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ NPEAD v2 (6-Axis Scoring) │ │ Uses BioBrain's structured evidence for all 6 axes │ └─────────────────────────────────────────────────────────────┘

Distillation Plan — Own the Intelligence

Model	Public Data	Architecture	Train Time	Inference
Binding affinity	PDBbind + BindingDB (5M points)	SchNet / GCN (3-layer)	4 hrs (A100)	5ms
ADMET multi-task	ToxCast + ChEMBL (500K)	Message-passing NN	2 hrs	3ms
Retrosynthesis	USPTO (1.8M reactions)	Template-based (Neuralsym)	6 hrs	20ms
Pocket classifier	sc-PDB (17K pockets)	3D CNN voxelized	1 hr	8ms
Target relevance	OpenTargets + DepMap (50K pairs)	XGBoost	10 min	<1ms
Molecular prior	ZINC + ChEMBL (10M SMILES)	Transformer (6-layer, 10M params)	12 hrs	15ms/mol

Total training: ~1 day on single A100 ($30-50 spot). Then inference at zero marginal cost inside pipeline.

Brain-Bus Messages

Direction	Message	Trigger
→ BioBrain	HYPOTHESIS_CREATED	Cortex hypothesis engine
→ BioBrain	TARGET_SELECTED	physician-rs discovery
→ BioBrain	VQE_RESULT_RECEIVED	Quantum brain
→ BioBrain	FAILURE_ATTRIBUTED	Outcome recording
BioBrain →	MOLECULE_GENERATED	MolGen complete
BioBrain →	BINDING_SCORE_CREATED	BindNet scored
BioBrain →	CANDIDATE_REJECT_RECOMMENDED	SafetyNet flag
BioBrain →	VQE_ESCALATION_REQUESTED	BindNet uncertain
BioBrain →	BIOMOLECULAR_OPINION_READY	Orchestrator complete

MindNetQ Inc. — Biomolecular Brain is Cortex's owned intelligence layer. Not one giant model — a specialist stack connected through the Brain-Bus.

🧬 Biomolecular Brain

Why BioBrain Matters

Module Status

Distillation Plan — Own the Intelligence

Implementation Phases

Live Pipeline Stats

Brain-Bus Messages

📋 BioBrain Progress Log