← Research

Drug × Protein

When the treatment hits the target — CYP competition meets structural coupling
JIM’S OVERSIMPLIFICATION

A drug works by coupling to a protein at a specific site. Stronger coupling means a more effective drug. But if it couples too strongly, you can’t get it off when you need to. The sweet spot is tight enough to work, loose enough to leave.

THE PROBLEM NOBODY CONNECTS

Doctors prescribe drugs to fix proteins. But the drugs themselves compete for enzymes. These are two separate conversations that happen in two separate departments, and they almost never talk to each other. The drug interaction model knows which drugs fight over enzymes. The protein model knows which mutations need fixing. Put them together and you see the whole patient for the first time.

THE CASE THAT PROVES IT

A breast cancer patient is on tamoxifen. Tamoxifen is a prodrug — it needs an enzyme called CYP2D6 to activate it. Without CYP2D6, tamoxifen is inert. Paper weight. The patient gets depressed (understandably). Doctor prescribes fluoxetine (Prozac). Fluoxetine is a strong CYP2D6 inhibitor. It blocks the exact enzyme tamoxifen needs. The cancer drug stops working. The tumor keeps growing. Nobody notices because two different doctors prescribed two perfectly reasonable drugs that happen to cancel each other out.

This is published. This is known. This still happens. A system that checks both layers — enzyme competition AND protein targets — catches this in milliseconds.

THE BIGGER PICTURE

The drug interaction engine and the protein engine are different views of the same patient. CYP3A4 is a bottleneck — 30 out of 49 drugs pass through it. That is a coupling bottleneck, the same math as everywhere else on this site. A patient is not a protein. A patient is not a drug interaction. A patient is both.

THE CONNECTION

The drug interaction engine models CYP enzyme competition — which drugs block each other's metabolism. The mutation scanner models structural coupling in disease proteins. These are different parts of the same patient.

THE CASE: Cancer patient on tamoxifen (CYP2D6 prodrug).
Doctor prescribes fluoxetine for depression (strong CYP2D6 inhibitor).

Fluoxetine blocks CYP2D6 → tamoxifen can't convert to endoxifen
NO active drug reaches the tumor → treatment fails.

The drug interaction kills the protein intervention.
Our two tools model each half. Together: the whole patient.

THE CROSS-PREDICTION

For each of our 5 disease proteins, what drugs should a patient AVOID because they compete for the same CYP enzymes as potential treatments?

Aβ42 (Alzheimer's):
  Treatments metabolized by CYP3A4 (donepezil, galantamine)
  Avoid: ketoconazole, erythromycin, grapefruit — CYP3A4 inhibitors

IAPP (Type 2 diabetes):
  Metformin is renally cleared (no CYP competition)
  But sulfonylureas use CYP2C9 — avoid fluconazole, amiodarone

α-synuclein (Parkinson's):
  Levodopa uses COMT/MAO, not CYP. Less CYP risk.
  But: entacapone inhibits CYP2C9 — watch warfarin coadministration

FUS / TDP-43 (ALS):
  Riluzole metabolized by CYP1A2
  Avoid: fluvoxamine (CYP1A2 inhibitor → riluzole toxicity)
  Avoid: smoking cessation (CYP1A2 inducer → reduced riluzole levels)

THE UNIFIED MODEL

In K terms: the drug interaction engine measures coupling between drugs at the metabolic layer. The protein engine measures coupling within the target at the structural layer. A full patient model needs both layers connected. The CYP bottleneck (CYP3A4 handles 30/49 drugs in our database) is itself a coupling bottleneck — the same Fiedler math from spectral forensics.

HONEST LIMITS

What this is: A known problem in pharmacogenomics. Drug-drug interactions
affecting treatment efficacy is well-studied (FDA labels, PharmGKB).

What we add: We show the K framework connects our two existing tools
(CYP modeling + FoldWatch protein analysis) into a unified patient model.
The connection is the contribution, not the pharmacology.
GUMPResearch · Support · [email protected] · terms