Seattle’s Talus Bio says it has a way around the bottleneck that keeps some of medicine’s most sought-after protein targets out of reach. Its new model, Ptarmigan-1, sidesteps structure prediction entirely – a shortcut the company pitches as a path to faster drug development for diseases that include cancer.
Unveiled on October 1, Ptarmigan-1 is described by Talus as the first structure-free model built to search the whole proteome. Instead of guessing a protein’s 3D shape, it learns from mass spectrometry data that records where a compound latches onto a target inside a living cell.
The focus is transcription factors, the proteins that decide which genes a cell switches on. They drive many diseases, cancer among them, yet drugmakers long wrote them off as undruggable because they rarely hold one shape: their flexible stretches keep moving, so a structure-based tool finds no stable cavity to aim at.
“If you try to fold a transcription factor, you just get a plate of spaghetti,” said Lindsay Pino, the company’s chief technology officer. “You can’t do drug discovery on a plate of spaghetti.”
Talus searched a library of 3.4 billion compounds against more than 20,000 human proteins. Pino puts the scale of the problem at about half the human proteome – proteins with no stable fold for a structure predictor to work from. A Talus preprint posted in July adds that, across some 20,000 human proteins, 87 percent still attract neither an approved medicine nor a potent small-molecule ligand.
On a single Nvidia H100 GPU, Ptarmigan-1 averaged about 10 milliseconds per compound, against 54 seconds for Boltz-2, an open-source structure model – a speedup of roughly 5,000-fold.
