CortexQ | Graph Intelligence

51,000+ edges across 5 graph layers. CortexQ reasons across molecular → protein → pathway → clinical → knowledge dimensions.

Graph-of-Graphs Architecture

┌─────────────────────────────────────────────────────────────────────┐ │ CortexQ Multi-Layer Graph │ ├─────────────────────────────────────────────────────────────────────┤ │ │ │ Layer 1: MOLECULAR (~46,500 edges) │ │ ┌─────────────────────────────────────────────────────────┐ │ │ │ Compound ──inhibits──→ Target (IC50/Ki/Kd, pChEMBL≥5) │ │ │ │ 50 oncology targets × 1000+ compounds each │ │ │ │ Source: ChEMBL API · Updated: weekly │ │ │ └─────────────────────────────────────────────────────────┘ │ │ ↕ cross-layer │ │ Layer 2: PROTEIN (~330 edges) │ │ ┌─────────────────────────────────────────────────────────┐ │ │ │ Protein ──interacts_with──→ Protein (STRING PPI) │ │ │ │ Co-expression, physical binding, pathway co-membership │ │ │ └─────────────────────────────────────────────────────────┘ │ │ ↕ cross-layer │ │ Layer 3: PATHWAY (2,763 edges) │ │ ┌─────────────────────────────────────────────────────────┐ │ │ │ Gene ──activates/inhibits──→ Gene (DIRECTIONAL) │ │ │ │ 14 curated cancer pathways (211 edges) + KEGG (2,552) │ │ │ │ Feedback loops, crosstalk, upstream/downstream │ │ │ └─────────────────────────────────────────────────────────┘ │ │ ↕ cross-layer │ │ Layer 4: CLINICAL (~100 edges) │ │ ┌─────────────────────────────────────────────────────────┐ │ │ │ Gene ──mutated_in──→ Cancer type │ │ │ │ Drug ──treats──→ Indication │ │ │ │ Biomarker ──biomarker_for──→ Condition │ │ │ └─────────────────────────────────────────────────────────┘ │ │ ↕ cross-layer │ │ Layer 5: KNOWLEDGE (~1,530 edges) │ │ ┌─────────────────────────────────────────────────────────┐ │ │ │ Gene ──resistant_to──→ Mechanism │ │ │ │ Gene ──participates_in──→ Pathway │ │ │ │ Source: heroes, PubMed NER, ClinicalTrials.gov │ │ │ └─────────────────────────────────────────────────────────┘ │ │ │ └─────────────────────────────────────────────────────────────────────┘

Edge Types

Each edge carries: source, target, type, weight (0-1), layer, evidence provenance, timestamp.

Edge Type	Badge	Weight	Meaning	Count
inhibits	inhibits	1.0	Drug/gene suppresses target activity	~46,500
activates	activates	1.0	Upstream signal turns on downstream	~80
interacts_with	interacts	0.9	Protein-protein physical interaction	~330
participates_in	pathway	0.7	Gene is member of this pathway	~2,552
feedback_activates	feedback	0.8	Negative/positive feedback loop	~15
resistant_to	resistant	0.8	Resistance mechanism connection	~200
downstream_of	downstream	0.7	Pathway cascade direction	~50
upstream_of	upstream	0.7	Pathway cascade direction	~50
mutated_in	mutated	0.8	Gene mutated in this cancer	~1,000
synthetic_lethal_with	SL	0.9	Co-essential pair (BRCA/PARP)	~20
sensitizes_to	sensitize	0.8	Combination synergy	~50
bypasses	bypass	0.8	Resistance escape route	~30

14 Curated Cancer Signaling Pathways

Hand-built directional edges encoding the biology that databases miss. Feedback loops, crosstalk, immune interactions.

Pathway	Edges	Key Signal Chain	Cancer Relevance
RAS/MAPK	27	EGFR→SOS1→KRAS→BRAF→MEK→ERK→MYC	Driver in 40% CRC, NSCLC
PI3K/AKT/mTOR	20	PIK3CA→AKT→mTOR + PTEN brake + AMPK opposing	Survival signaling, PTEN loss
WNT/β-catenin	16	APC destruction→CTNNB1→TCF→MYC	CRC initiating event (APC loss)
TGF-β/SMAD	14	TGF-β→TGFBR→SMAD→EMT + immune exclusion	Stromal barrier, T cell exclusion
cGAS/STING	14	cGAS→STING→TBK1→IRF3→IFN→CXCL9/10	Innate sensing, T cell recruitment
PD-1/Checkpoint	11	PD-L1⊣PD-1⊣TCR + MAPK→PD-L1	Immune evasion, checkpoint therapy
Cell Cycle	18	CCND1→CDK4/6⊣RB⊣E2F→CCNE→CDK2	CDK4/6 inhibitors, RB loss
DNA Damage	14	ATM/ATR→CHEK→p53→BRCA→RAD51	PARP inhibitors, HRD
Apoptosis	18	BCL2⊣BAX→cytochrome c→caspase cascade	Venetoclax, MCL1 resistance
Myeloid Suppression	12	CSF1→CSF1R→STAT3→ARG1/IDO1⊣CD8	MSS CRC immune desert, TAMs
AR Signaling	11	AR→KLK3/TMPRSS2 + AKT/MAPK cross-activation	Prostate, castration resistance
Hippo/YAP	12	LATS⊣YAP→TEAD→CTGF + KRAS→YAP bypass	Drug resistance, bypass route
Notch	9	DLL/JAG→NOTCH→HES1 + MYC	Stemness, lineage decisions
JAK/STAT	15	IFN/IL6→JAK→STAT1/3→immune + survival	Immune signaling, inflammation

Multi-Hop Reasoning Examples

Data Sources & Refresh

Storage & Access

CIFOG GCN — Graph Neural Network

How the Graph Grows Autonomously

Question	Hops	Layers Crossed	Answer Path
"If we block KRAS, what reactivates?"	4	Pathway	KRAS→BRAF→MEK→ERK→[feedback]→SOS1→KRAS
"What compounds hit PI3K pathway?"	2	Molecular + Pathway	Compound──inhibits→PIK3CA──participates_in→PI3K/AKT
"Why does anti-PD1 fail in MSS CRC?"	3	Pathway + Knowledge	TGF-β⊣CD8 + CSF1R→ARG1⊣CD8 + no CXCL10 (STING off)
"Is there a drug for BRAF's bypass route?"	3	Pathway + Molecular	BRAF──activates→MEK + KRAS──activates→PIK3CA ← [compounds exist]
"What's the combination for TBK1?"	2	Pathway + Clinical	TBK1──in→cGAS/STING + PD-1──checkpoint→ = TBK1i + anti-PD1

Source	Edges	Refresh	Method
ChEMBL API	~46,500	Weekly (Sunday 4am)	REST API → 50 targets × paginated activities
KEGG REST API	2,552	Weekly	14 cancer pathway gene memberships
Curated Pathways	211	Manual	Hand-built directional signaling (pathway_enrichment.py)
Internal Heroes	~1,530	Weekly	DDB scan → gene→cancer, gene→pathway, gene→resistance
STRING PPI	~330	On seed	Protein-protein interactions (pre-loaded)
PubMed NER	Growing	Weekly	Literature mining → gene-gene relationships (graph_enrichment.py)

Component	Location	Access Pattern
Edges	DynamoDB `cifog-edges`	pk=source#target, sk=edge_type:source_id
Nodes	DynamoDB `cifog-nodes`	pk=gene, attributes: authority, vector, edge_count
Vectors (256-dim)	Qdrant Cloud (free tier)	Cosine similarity search for CIFOG-GCN
GCN Model	S3 (trained nightly)	3-layer: 286→128→64→1 authority regression

CortexQ is the master. The graph is CortexQ's memory of how things connect.
MindNetQ Inc. · CIFOG Graph Intelligence · Updated July 11, 2026

📊 Graph Intelligence — CIFOG Multi-Layer