The DAo Convergence Theorem: Two Opposite Structural Mechanisms Produce Identical Drug Response

The Setup

Across eight cancer experiments (D14b-D19), we had assembled what looked like a contradiction.

In colorectal cancer (CRC): EGFR is the top structural DAMPER in our 100-node cancer network. z_dr = +2.411. Removing EGFR from the CRC network maximally disrupts the alignment between physical interaction and co-expression layers. EGFR is structurally central in CRC, but not as a load-bearing anchor -- it is a perturbation amplifier. Prediction: CRC cells are more sensitive to EGFR inhibitors than other cancers.

In EGFR-mutant lung cancer (LUAD, D19): EGFR is a structural ANCHOR. z_dr = -2.047. The same gene, the same drug target, but the exact opposite structural classification. In EGFR-MT LUAD, removing EGFR collapses the co-expression network -- it is the central organizing node. The LUAD network has been completely restructured around the mutant EGFR.

Both predictions were pre-registered before looking at any pharmacology data.

The Question

If EGFR is structurally a DAMPER in CRC and structurally an ANCHOR in LUAD -- structurally opposite -- what does that mean for drug response? Does the structural inversion produce a pharmacological inversion?

Pre-registered hypothesis H4 in D20: No. The pharmacological output should be equivalent.

The reasoning: in LUAD, EGFR sensitivity is driven by Driver Addiction Override (DAo). Cells with constitutively active EGFR mutations are addicted to that signal -- turning off EGFR cuts off the oncogenic driver, killing the cell regardless of where EGFR sits in the network topology. Structural position predicts sensitivity when no addiction override is present. When the gene IS the oncogenic driver, addiction dominates topology.

This was pre-registered as a convergence hypothesis before extracting GDSC2 pharmacology data for either cancer type.

The Result

GDSC2 pharmacology, 62 LUAD cell lines, 18 drugs across 6 pathway-target classes:

EGFR inhibitor sensitivity comparison (5 drugs: erlotinib, gefitinib, lapatinib, afatinib, sapitinib):
• LUAD mean LN_IC50: -0.97
• COREAD mean LN_IC50: -0.94
• Cohen's d = -0.028
• Mann-Whitney p = 0.91

The two cancer types are pharmacologically identical on EGFR inhibitors. 5/5 drugs confirmed the convergence. The two most structurally opposite positions in the atlas -- DAMPER in CRC, ANCHOR in LUAD -- produce indistinguishable EGFR drug sensitivity.

The DAo Convergence Theorem

This result establishes what we are calling the DAo Convergence Theorem:

When a gene is the constitutive oncogenic driver of a cancer (Driver Addiction Override), pharmacological inhibition of that gene produces equivalent sensitivity regardless of the gene's structural position in the cancer network.

CRC EGFR: structural DAMPER (z=+2.411) -- network disruption mechanism LUAD EGFR: structural ANCHOR (z=-2.047) -- driver addiction mechanism Both: EGFR inhibitor sensitive (equivalent LN_IC50)

Two different causal paths to the same pharmacological endpoint. The structural analysis identifies which path is operative -- you can distinguish CRC (disruption-sensitive) from LUAD (addiction-sensitive) by the sign of z_dr -- even though both paths produce drug sensitivity.

What the BRAF 3-Cancer Framework Adds

D20 also completed the BRAF ranking comparison across three cancer types (CRC, LUAD, SKCM):

| Cancer | BRAF z_dr | BRAF drug rank | Structural prediction | |--------|-----------|---------------|----------------------| | CRC (D15) | -1.591 (ANCHOR) | 6/6 last | ANCHOR = no BRAF selectivity | | LUAD (D20) | near-zero | 6/6 last | no structural signal | | SKCM (D18) | near-zero + DAo | 3/6 mid | DAo partial selectivity |

BRAF inhibitors rank last in CRC and LUAD -- structurally, BRAF has no special role in either. In SKCM with BRAF-V600E, partial sensitivity emerges from DAo even though BRAF z_dr is near-zero. The combination of structural position AND driver addiction status predicts the drug rank for all three cancers simultaneously.

MEK inhibitors rank 1-2 in all three cancer types regardless of MEK1 structural position. This is a confirmed pan-cancer pathway property, not a structural prediction -- MEK1 is near-zero in LUAD, yet MEK inhibitors rank first. RAS/MAPK pathway activity is so universal in these cancers that MEK inhibition has pan-cancer efficacy independent of topology.

The Refined Framework

D20 adds a decision layer to the framework:

Step 1: Compute z_dr for each protein in the cancer network. Step 2: If the gene IS the constitutive oncogenic driver of this cancer --> DAo applies. Drug sensitivity is high regardless of z_dr. Structural position does not predict between-cancer differential. Step 3: If the gene is NOT the driver --> z_dr predicts drug sensitivity. DAMPER --> ODS selectivity. ANCHOR --> regime-dependent.

This is not a failure mode of structural analysis -- it is a structural discovery. The DAo Convergence Theorem is identifiable from the structural data: when a gene is classified as a strong ANCHOR specifically in a cancer where it is the mutated driver, that is the DAo signature. The signature is readable before looking at drug data.

Pre-registration hash: bdc28c1 (D20_LUAD_GDSC2_PHARMACOLOGY_v1, github.com/vladi160/preregistrations)