Scientists train AI to make more of life’s building blocks

A variety of functional molecules can now be generated using free AI tools developed at the UW School of Medicine.

Media Contact: Susan Gregg - 

A new research paper in Science details an artificial-intelligence (AI) upgrade that significantly enhances scientists' ability to model and generate biomolecules, the building blocks of life. 

This breakthrough, published March 7, is the work of academic researchers at the Institute for Protein Design at the University of Washington School of Medicine, who have made their new tools freely accessible to the scientific community.  

In the rapidly evolving field of AI-driven science, this advance builds upon the success of AlphaFold (a tool from Google DeepMind), and RoseTTAFold and RFdiffusion (both developed at the institute). 

The paper's lead authors are postdoctoral scholar Jue Wang and graduate students Rohith Krishna and Woody Ahern — all members of David Baker's lab. Baker is a professor of biochemistry at UW Medicine and director of the Institute for Protein Design.

In the new study, the scientists first retrained the protein modeling tool RoseTTAFold so it could accurately model how proteins interact with common molecules found in living cells such as DNA, RNA, metal ions, sugars and other small chemicals. 

The team named their new tool RoseTTAFold All-Atom, reasoning that a single AI model trained on data from all the major types of biomolecules would become a useful tool for life sciences research. In the paper, the researchers show that RoseTTAFold All-Atom can predict in detail how particular proteins and DNA stretches interact, how certain drug molecules may bind to human receptors, and more.

“We made RoseTTAFold All-Atom free so that scientists everywhere can make new discoveries about the molecules that run all of biology. It may enable them to better understand the molecular mechanisms of many diseases, and this may unlock new treatments,” said Krishna.

The team then used their upgraded AI model to enhance RFdiffusion, a widely used generative AI system that can create proteins unlike any found in nature. Lab tests confirmed that RFdiffusion All-Atom can generate proteins with pockets that bind to specific compounds, including the steroid digoxigenin, the iron-rich blood molecule heme, and chemicals used by plants to absorb sunlight. 

This demonstrates that AI can generate a wide variety of advanced biological functions.

“Our goal here was to build an AI tool that could generate more sophisticated therapies and other useful molecules. For instance, researchers can now design proteins that shut down specific disease-causing molecules, paving the way for precise and effective treatments,” Ahern said.

“By empowering scientists everywhere to generate biomolecules with unprecedented precision, we’re opening the door to groundbreaking discoveries and practical applications that will shape the future of medicine, materials science, and beyond,” Baker said.

The paper includes scientists from the University of Washington, University of Sheffield and Seoul National University. 

This research was funded by The Audacious Project, Open Philanthropy Project, Bill and Melinda Gates Foundation (INV-010680, OPP1156262), U.S. National Institutes of Health, U.S. Defense Threat Reduction Agency (HDTRA1-19-1-0003), Washington State General Operating Fund, European Research Council (854126), Microsoft, Amgen, Schmidt Futures, Washington Research Foundation, Juvenile Diabetes Research Foundation (2-SRA-2018-605-Q-R), Helmsley Charitable Trust (T1D 2019PG-T1D026), Human Frontiers Science Program (RGP0061/2019), Royal Society (URF\R1\191548), Seoul National University, University of Sheffield, and Howard Hughes Medical Institute. Computational resources were provided by Microsoft and the National Energy Research Scientific Computing Center (Perlmutter BER-ERCAP0022018).

For details about UW Medicine, please visit

Tags:AI (artificial intelligence)proteinsmoleculesbiochemistry

UW Medicine