John Taylor, Professor of Economics at Stanford University and developer of the "Taylor Rule" for setting interest rates | Stanford University
John Taylor, Professor of Economics at Stanford University and developer of the "Taylor Rule" for setting interest rates | Stanford University
Stanford scientists have developed a new machine learning-based method to more quickly and accurately predict the molecular changes that will lead to better antibody drugs. Published in Science on July 4, the approach combines the 3D structure of the protein backbone with large language models based on amino acid sequence, allowing researchers to find, in minutes, rare and desirable mutations that would otherwise only be found with exhaustive experiments.
Led by Peter S. Kim, professor of biochemistry and institute scholar at Sarafan ChEM-H, and Brian Hie, assistant professor of chemical engineering, the team showed that they could improve a once FDA-approved SARS-CoV-2 antibody that had been discontinued due to its ineffectiveness against a new strain in November 2022. Their approach resulted in a 25-fold improvement against the virus.
“A lot of effort in AI and drug development is centered around amassing tons of data about how well a certain molecule performs a certain task so that a computer can learn enough to design a better version,” said Kim. “What’s remarkable is that we’ve shown that structure can be used in lieu of a lot of that data, and the computer will still learn.”
“Now, more antibodies actually get a shot at being optimized,” said Hie, who is also an innovation investigator at the Arc Institute.
When faced with the challenge of finding the best amino acid sequence, scientists often make millions and test them in miniaturized versions of biological systems. They hope that the best drug in a dish will also be effective in humans.
“It’s a lot of guess and check,” said Hie. “The goal of many intelligent algorithms is to remove the guesswork from this.”
To speed up the process, scientists have developed ChatGPT-like machine learning algorithms trained on amino acid sequences of millions of proteins to predict desirable mutations. These models often point scientists toward sequences that are unstable or worse than where they started because protein function depends not only on amino acid sequence but also on its 3D structure.
The key to developing a better prediction algorithm was structure. The team constrained possible beneficial mutations determined by sequence-based models to those preserving the starting protein's 3D shape. In December 2022, they tested it on a recently discontinued SARS-CoV-2 antibody therapy.
“The prevailing theory was that trying to improve this antibody would fail,” said Varun Shanker, medical student and graduate student in biophysics who led the study. “The virus evolved as it spread through millions of people to avoid these antibodies.”
Using purely sequence-based models resulted in modest improvements; however, their structure-guided approach saw significant increases.
“We were finally catching up to the virus,” said Shanker.
Most efforts using AI for drug development rely on generating vast amounts of data about unique protein sequences' function and performance—a time-consuming process resulting in models tailored for specific tasks. This model does not require input about what the protein does or any lab experiments; instead, it uses protein coordinates as proxies for performance.
Early experiments indicate this approach is generalizable to other proteins like enzymes which catalyze chemical reactions within our bodies. Researchers found half of tens pointed proteins were better than their starting points.
This tool could respond quickly to emerging diseases while lowering barriers for creating effective medicines requiring lower doses benefiting more patients. For diseases like HIV where infrequent large doses protect from infection—this could be transformational.
The team has made their model and code freely available.
“This is an exciting example of deep learning democratizing building better proteins,” said Shanker. “It allows people to develop new medicines and opens new scientific exploration areas previously inaccessible.”
Theodora Bruun co-authored this study alongside Kim who holds memberships across various research institutes including Bio-X and Wu Tsai Neurosciences Institute; Hie belongs likewise plus holds faculty fellowship at Stanford Data Science; Shanker participates within Stanford Medical Scientist Training Program supported by Virginia & D.K Ludwig Fund for Cancer Research along Chan Zuckerberg Biohub support.
Rebecca McClellan: rmcclell@stanford.edu
Alerts Sign-up