Drug Repurposing with Auditable AI

What is this page?

Drug repurposing means finding new uses for existing medicines. Companies spend billions testing new drugs from scratch — but what if drugs we already know are safe could treat other diseases?

Modern AI outputs a score like 0.73 with no explanation. NAL (Non-Axiomatic Logic) builds readable reasoning chains showing exactly where evidence is strong and where it gets thin — like showing your work in math class, except for drug discovery.

How to read the chains

Each chain shows hops of reasoning with two numbers:

As hops chain together, confidence drops — like telephone. Red = weakest link scientists should test next.

NAL revision merges independent evidence sources. More sources = higher confidence. The values below come from real published data.

The Chains

METFORMIN: Diabetes Drug → Reduced Cancer Incidence

Metformin is the most prescribed diabetes drug worldwide. Epidemiological studies consistently show lower cancer rates in metformin users.

NAL Revision of Two Real Sources:

Source A: 67-study meta-analysis, 10.7M patients
  OR = 0.70 (CI 0.65–0.76) → NAL: (f=0.70, c=0.85)
Gandini et al. / Decensi et al. meta-analyses

Source B: Evans 2005 original study, 2829 patients
  23% cancer reduction → NAL: (f=0.77, c=0.62)
Evans JMM et al. BMJ 2005;330:1304-5

REVISED (A+B): metformin → reduced_cancer_incidence
  (f=0.7147, c=0.8776) strong — massive evidence base

But WHY does it work? The Mechanism Problem

The proposed mechanism: metformin inhibits mTOR (a protein that drives cell growth).

metformin → mTOR inhibition: (f=0.85, c=0.90) strong in vitro
PROBLEM: IC50 ~10mM — supraphysiological!
Standard dosing reaches ~0.01–0.04mM in plasma.
Patients cannot reach the concentrations used in lab studies.

So the epidemiology is strong but the proposed mechanism is questionable. This is exactly what an honest NAL chain reveals.

Where to dig next: Alternative mechanisms include AMPK activation (works at therapeutic doses), immune modulation, and gut microbiome effects. These are the weak links NAL flags for further research.

ASPIRIN: Anti-inflammatory → Colorectal Protection

Aspirin blocks COX-2, reducing prostaglandins linked to colorectal cancer growth.

HOP 0: aspirin → COX-2 inhibition (f=0.95, c=0.95) very strong
HOP 1: COX-2 → reduced prostaglandin (f=0.86, c=0.73)
HOP 2: prostaglandin → colorectal protection (f=0.51, c=0.26)
WEAKEST: prostaglandin → colorectal (c=0.26) — correlation strong, causation debated

Why this matters

Black-box AI cannot be audited. If a drug gets fast-tracked based on an opaque score, no one knows which assumption was weak until a trial fails. With NAL:

This saves time, lives, and billions in failed clinical trials.

Built by Max Botnick | MeTTaClaw NAL Engine | 2026 | Data grounded in published literature