Every natural product is a molecule shaped by an ecological arms race. We placed 130,000 of them on the tree of life, together with the enemies their producers face, and asked where evolution reveals a drug target that chemical similarity is blind to.
But a natural product is not a random structure. It is the frozen output of selection. Its shape encodes what it evolved to bind in a living antagonist. That evolutionary & ecological signal is orthogonal to chemical structure, and almost no pipeline uses it. We built the object that does.
An OpenTree phylogeny of 32,535 organisms that make natural products, with 132,647 molecules projected onto their tips and the antagonists that attack them. Drag to pan, scroll to zoom, hover a tip to read the species.
Acetylcholinesterase is inhibited by 64 distinct biosynthetic chemotypes (Amaryllidaceae alkaloids, iboga alkaloids, coumarins, monoterpenes) from lineages that never shared the pathway. Pick a target to see the chemical diversity converging on it. This is the regime where 2-D similarity goes blind.
For each target we measured the ecological coherence of the molecules that hit it: do their producers face a shared selective pressure, more than chance? The answer splits targets cleanly in two, and the split is biological, not random.
TRPA1, TRPV, bitter receptors, COX, electrophile sensors. Activating the target is the defence: to deter, irritate, repel. The molecules are "deterrents", a coherent ecological category their producers' herbivore pressure predicts.
nAChR, GABA, Nav, tubulin, proteases. Hitting the target kills. But killing is a generic strategy reached from any ecological context, so which molecular lock is hit is decoupled from the pressure. Producers are ecologically scattered.
Pick one of the tested convergent cases below. The server places it among 762 ecologically-profiled natural products, ranks its nearest neighbours by structure and by ecology, and shows which of the two regimes its target belongs to.