Finding new medicines hidden in old ones.
A search for forgotten cures: screening 100 approved or clinical-stage drugs against the proteins behind dengue and other neglected tropical diseases, using physics, chemistry, and a measured dose of machine learning.
A billion people, and almost no new drugs.
Neglected tropical diseases like dengue, chikungunya, and leptospirosis affect more than a billion people, and Indonesia carries a disproportionate share of that burden.
Yet building a brand-new drug takes well over a decade and a billion dollars. For diseases that mostly affect the world's poorest, that investment rarely comes. The pipeline is, in practice, empty.
What if the cure already exists?
Drug repurposing asks a simple question: instead of inventing a new molecule, could a medicine we already trust, already proven safe in millions of people, work against a disease it was never designed for?
It is faster, cheaper, and safer at the starting line, because the hardest question, "is this safe for humans," already has an answer. GeneTropica screens a curated library of approved or clinical-stage drugs to find those hidden second lives.
Six stages, from molecule to candidate.
A curated library
100 approved or clinical-stage drugs, chosen to span chemical space, become the search pool. 99 docked successfully; auranofin contains a gold atom that AutoDock Vina cannot model.
Six disease targets
Key proteins from dengue, chikungunya, and leptospirosis, the machinery each disease depends on to survive.
594 docking runs
AutoDock Vina computes how snugly every drug fits into every target (99 drugs × 6 targets), like testing keys against a lock.
Two honest rankings
Candidates are scored both by raw binding strength and by ligand efficiency, the binding earned per atom, shown side by side.
A machine-learning prior
A model trained on known actives adds a hint of likely activity. It is treated as a clue, never the verdict.
Molecular dynamics
The strongest candidates are stress-tested with 50-nanosecond simulations to see whether the fit actually holds.
These are real computed results, not illustrations. Every run was executed on Google Colab: the 50-nanosecond molecular-dynamics simulations on an NVIDIA A100 GPU, and the docking, machine learning, and conservation analysis on CPUs.