University of Washington Team Builds Open Alternative to DeepMind's Protein AI
After DeepMind delays releasing its AlphaFold2 code, a University of Washington lab creates an accessible open-source alternative called RoseTTAFold. The tool allows biologists to predict protein structures without needing advanced computational skills.
DeepMind makes a massive leap in biology by solving the protein structure prediction problem with its AlphaFold2 neural network. This breakthrough divides the field into two distinct eras, as the AI system successfully determines what proteins look like based on their amino acid sequences. Understanding these structures helps scientists research cellular functions and accelerates the drug discovery process.
Despite the scientific significance of this achievement, DeepMind does not promptly share the underlying code with the public. This seven-month delay prompts a team at the University of Washington, led by David Baker, to develop their own prediction model called RoseTTAFold. The team releases this alternative a full month before DeepMind finally publishes its official manuscript and code.
While RoseTTAFold does not quite match the peak performance of AlphaFold2, it immediately becomes the most successful algorithm that scientists can actually use. The University of Washington team prioritizes accessibility by building a simple submission tool that allows researchers to predict protein structures without needing advanced computational expertise.