“Chai Discovery believed that the adoption of diffusion models, which can sample multiple distinct hypotheses instead of averaging them, would be a key driver for improving performance in protein structure prediction.”