Orbital Labs Cannot Solve Drug Discovery's Real Problem

A British startup just launched a laboratory into orbit to grow protein crystals in microgravity, betting that better structures will crack the code on aging diseases.

The company promises that AI trained on these orbital datasets will accelerate drug discovery for Alzheimer's, cancer. The rest of the longevity market. It sounds like the future.

In 2006, the International Space Station had been running a protein crystal growth program for nearly a decade. NASA and the ESA had invested heavily in the assumption that microgravity would solve a concrete problem. Crystals grown in orbit were cleaner, larger, more ordered than terrestrial versions. With better crystals came better X-ray diffraction data, better data meant better protein structures. Better structures meant faster drug discovery.

The program produced beautiful structures. It also produced almost no drugs. By 2010, reviewing the clinical outcomes, researchers found the bottleneck had never been crystal quality at all. It had always been what happened after you had the structure. The actual work of drug discovery lives downstream, in the computational slog and chemical intuition required to design a molecule that binds to that protein in precisely the right way, in the right tissues, without poisoning everything else.

A perfect crystal is necessary but not sufficient.

The British venture is built on the assumption that AI will change this equation. Feed machine learning models enough protein structures, the theory goes. The models will learn the binding rules that human chemists still struggle to intuit. But the models themselves have to be trained on experimental data about what actually works in living systems. Structures from orbit are interesting, data about whether a compound fails in a mouse or succeeds in a trial is what teaches the algorithm what matters. You can't generate that in microgravity.

What's different now is not the science. It's the venture capital model. Orbital labs are cheaper to operate than they were in 2006. That means the startup can fail more cheaply. It also means we'll get publishable structures without clinical application, and everyone will move on.