Cancer is a network disease. Mutations don't just break one protein — they cascade through interconnected signaling pathways. We modeled 9 major cancer pathways with 19 cross-pathway crosstalk edges from the literature, then computed K (coupling strength) for every node.
K = degree / max_degree in the combined network. The highest-K nodes are the ones whose knockout disrupts the most downstream signaling.
MM10P warning: the hub ranking below is determined by our choice of crosstalk edges. We added 5 edges to CTNNB1, making it #1 by construction. In real protein interaction networks (STRING database), TP53 has ~6,000 partners vs CTNNB1's ~1,200. TP53 is the real master hub. Our ranking reflects our edge selection, not biology. This is a 90-node toy network — real PPI networks have thousands of nodes. The cascade patterns are illustrative, not predictive.
| Rank | Protein | K | Connections | Pathway |
|---|---|---|---|---|
| 1 | CTNNB1 | 1.000 | 10 | WNT |
| 2 | MTOR | 0.600 | 6 | PI3K/AKT |
| 3 | RBPJ | 0.600 | 6 | NOTCH |
| 4 | GSK3B | 0.600 | 6 | WNT |
| 5 | KRAS | 0.600 | 6 | RAS/MAPK |
| 6 | PIK3CA | 0.500 | 5 | PI3K/AKT |
| 7 | TP53 | 0.500 | 5 | p53 |
| 8 | AKT1 | 0.500 | 5 | PI3K/AKT |
| 9 | STAT3 | 0.500 | 5 | JAK/STAT |
| 10 | NOTCH1 | 0.500 | 5 | NOTCH |
CTNNB1 (β-catenin) is the master hub with 10 connections across 5 pathways via crosstalk. It receives signals from WNT, TGF-β, Hedgehog, and Hippo. This matches clinical data: WNT/β-catenin dysregulation appears in >90% of colorectal cancers.
When you knock out a high-K node, the signal loss propagates through the network. Impact decays with distance and K of downstream nodes. Three-step propagation:
| Knockout | K | Proteins Hit | Pathways | Cross-Pathway |
|---|---|---|---|---|
| CTNNB1 | 1.000 | 17 | 5 | 9 |
| GSK3B | 0.600 | 15 | 7 | 9 |
| PIK3CA | 0.500 | 14 | 5 | 8 |
| MTOR | 0.600 | 13 | 5 | 6 |
| STAT3 | 0.500 | 12 | 5 | 8 |
| NOTCH1 | 0.500 | 11 | 4 | 4 |
| KRAS | 0.600 | 10 | 4 | 5 |
| TP53 | 0.500 | 8 | 3 | 4 |
GSK3B knockout disrupts 7 of 9 pathways — it sits at the intersection of WNT, RAS/MAPK, and their downstream targets. This is why GSK3 inhibitors have broad anticancer activity but also broad toxicity.
| Cancer | Pathways | Hub Count | Kmax |
|---|---|---|---|
| Colorectal | 5 | 4 | 1.000 |
| Lung | 5 | 2 | 0.600 |
| Breast | 3 | 2 | 0.600 |
| Glioblastoma | 2 | 2 | 0.600 |
| Pancreatic | 3 | 1 | 0.600 |
| Melanoma | 2 | 3 | 1.000 |
Colorectal cancer spans the most pathways in our model (5), which is biologically correct — known drivers include APC, KRAS, TP53, PIK3CA, and SMAD4 across WNT, RAS, p53, PI3K, and TGF-β. However, the specific ranking (#1 most vulnerable) and hub counts depend on our 90-node network. Lung and breast cancer also span 5+ pathways in fuller networks. Colorectal is among the most multi-pathway cancers, not necessarily the most.
The network is built from KEGG pathway data with 19 hand-selected cross-pathway edges. K for each protein = its degree divided by the maximum degree in the combined network. Knockout propagation uses a 3-step breadth-first search with impact decay proportional to downstream K. No machine learning.
MM10P honest limits: this is a 90-node toy network. Real protein interaction databases (STRING, BioGRID) contain thousands of nodes with millions of edges. Our hub rankings are determined by which 19 crosstalk edges we chose to include. Different edge selections produce different rankings. The cascade patterns demonstrate the principle of network vulnerability but should not be used for clinical decisions. Next step: wire in real PPI interaction counts from STRING.
Computed in 1.2ms on Mac Mini M4. Pathway data: KEGG, Reactome. Cross-pathway edges: PubMed literature review.