Tipoca
Every cell in your body runs on a program written in four letters, translated into twenty amino acids and executed by molecular machines we are only beginning to understand. Biology is the most sophisticated technology on Earth. For most of history we have studied it the way astronomers once studied the sky: by looking, waiting, and writing down what we saw.
That is changing. Neural networks now predict the shape of a protein from its sequence about as well as years of laboratory work. Models have designed proteins that never existed in nature, and language models trained on genomes have begun to read a grammar no one ever wrote down.
Tipoca is named after a city on a storm-covered ocean world, a place where life was not waited for but designed. We think that is the next century of biology: not only discovery, but engineering.
The search problem
Biology's possibility space is absurdly large. A modest protein of a hundred amino acids can be written in 20100 ways, more than there are atoms in the observable universe. Evolution has explored a vanishing fraction of it. Experiments have explored even less.
Even a small, well-defined search is brutal. Take a library of a million variants of a single protein, and look for the handful that bind a target.
The bottleneck was never the pipette. It was knowing where to point it.
From letters to life
A sequence is not a function. Between the two sit layers of physics and context: a chain folds into a structure, structures meet and bind, interactions add up to the behaviour of a cell. Each layer has been studied on its own, by different fields, with different data.
We build models that connect them, so that what is learned on one layer sharpens the predictions on the next.
The loop
Models alone are not enough. The data biology needs most, what happens when you change something, mostly does not exist yet. So we generate it. Every design a model proposes is synthesized, measured and fed back. Prediction, experiment, correction.
The loop runs continuously, and every turn makes the next one cheaper.
The map
Every protein ever sequenced is a point in a space that models can learn. Related proteins gather into families, families into folds. Between them lie large dark regions: sequences that could exist, and might be useful, but that nothing alive has ever made.
That is where we are going.