Proteins are origami. Disease happens when the origami gets sticky and clumps, or when it is too floppy and wanders. The fix is almost comically simple: sticky things need electric charge to repel each other. Floppy things need grip to fold. Six diseases. Two rules. Same physics.
Every protein misfolding disease follows one of two failure modes. Either the protein has an oily patch that makes copies stick to each other (aggregation), or it has no structure at all and drifts into toxic clumps (disorder). The body uses one error pattern. The fix is one pattern too.
We screened over 4,500 mutations across six disease proteins. The answer is the same every time. Too sticky? Add charge. Charged residues repel like same-pole magnets. Too floppy? Add anchors. Hydrophobic residues create a structural core from nothing. The engine reads the protein and prescribes the opposite of what is wrong.
| Disease | Protein | Problem | Strategy | Best Hit | Effect |
|---|---|---|---|---|---|
| Alzheimer’s | Aβ42 | KLVFF sticky | Charge | V18D | ↓28% agg |
| Parkinson’s | α-syn | NAC oily | Charge | V70D | ↓22% agg |
| Type 2 diabetes | IAPP | NFGAIL sticky | Charge | L16K | ↓50% agg |
| Alzheimer’s (tau) | Tau | PHF6 sticky | Charge | I278D | Hotspot gone |
| ALS (FUS) | FUS LCD | 98% floppy | Anchors | T11V+T71V | Core formed |
| ALS (TDP-43) | TDP-43 | Half/half | Both | I151D | ↓11% agg |
A vertex decouples. Edges weaken. The shape loses coherence. This is aggregation.
Aβ42 has a stretch in the middle — five amino acids called KLVFF — that is extremely sticky. Oily. Hydrophobic. It finds other copies and sticks to them, face to face, until you have a plaque that kills neurons.
We screened 798 mutations in 1.4 seconds. The top 10 mutations that reduce stickiness? All add electric charge. Zero add more oil. Ten out of ten. A drug in Phase 3 trials — tramiprosate (ALZ-801) — does exactly this: carries a negative charge to the aggregation surface. We found the same answer from the math. Nobody told the engine about tramiprosate.
The Arctic mutation (E22G) proves it in reverse: removing charge near KLVFF causes Alzheimer’s in the 50s instead of the 70s. Our answer is literally the inverse of a known disease-causing mutation.
Alpha-synuclein has a greasy stretch (NAC region, residues 61-95) that clumps into Lewy bodies, killing dopamine neurons. The wild-type protein is already borderline — low risk, net charge -8.9. The famous A53T familial mutation barely moves the numbers. This protein is one nudge from disaster.
V70D — charge in the heart of the NAC core — drops aggregation 22%. 7 out of 10 top stabilizing mutations add charge. Zero add hydrophobic. No existing drug targets this position. Levodopa replaces lost dopamine but does nothing about the clumping.
IAPP clumps in the pancreas, killing beta cells. 90% of Type 2 diabetes patients have amyloid deposits at autopsy. L16K — positive charge at position 16 — cuts aggregation 50% and spontaneously creates 27% helix from zero. The charge does not just prevent clumping — it tells the protein how to fold.
Tau is the brain’s rebar. It holds microtubules together. When tau detaches and clumps (PHF6: VQIVYK at positions 274-280), the scaffolding collapses. Tau tangles correlate more strongly with cognitive decline than amyloid plaques do.
I278D — one charged amino acid in the middle of VQIVYK — eliminates the entire 7-residue aggregation hotspot. The single largest sticky stretch in the whole 441-amino-acid protein, gone with one change.
FUS is the opposite of every other disease on this page. It has no structure at all. 98% floppy noodle. No skeleton. No core. Without structure, it drifts into clumps that kill motor neurons. There is no FDA-approved drug.
For the sticky proteins, charge worked. For FUS, charge does nothing (explicitly tested). The answer is reversed: add hydrophobic anchors to give the floppy chain a reason to fold. 9 out of 10 top stabilizing mutations add hydrophobic residues. Two anchors (T11V + T71V) are the minimum effective dose — they create a structural core that never existed, while preserving the flexibility FUS needs for RNA processing. A brace, not a cast.
TDP-43 is half structured, half disordered. The structured half (RRM domains) breaks the Alzheimer’s way — charge at I151 reduces aggregation 11%. The floppy tail (LCD) breaks the FUS way — it probably needs hydrophobic anchors. TDP-43 is a protein with an identity crisis. It needs both strategies simultaneously.
Wherever a hydrophobic surface drives pathological aggregation, charge disrupts it. The engine finds the exact position. The physics does not change.
The engine does not know which rule to apply in advance. It reads the protein and figures it out from the math. Every time.
This is computational research, not medical advice. The engine identifies molecular strategies from sequence analysis. Clinical validation requires wet-lab experiments and regulatory approval. Drug matches are independent computational findings, not clinical endorsements.