Automation accelerates the search for stable proteins

Harnessing the power of robotics and artificial intelligence, researchers at Princeton Engineering and Rutgers University have found a way to design stable proteins in a fraction of the time of the current state-of-the-art. The team’s robotics platform speeds things up more than tenfold, and its computational approach finds solutions anywhere from weeks to years faster than is possible with human intelligence alone.

Protein stabilization is a central challenge for research on drug creation, biofuel production, and plastic recycling. Scientists now use their knowledge of chemistry to estimate which chemical compounds will combine well with proteins under different conditions. The conventional approach uses trial and error to refine the results. This painstaking method can take months as scientists create and test molecule samples, and often leads nowhere.

In the new system, engineers use machine learning model to identify chemical compounds it is more likely to stabilize desired proteins. The model helps narrow hundreds of thousands of possibilities down to a few likely candidates. A robotic assembly platform produces samples of the molecules for evaluation. Combining the robotic platform with the machine learning model results in just a few days.

This twin-turbo approach offers an additional advantage: Due to its ability to process large amounts of data, the machine learning model often recommends candidate molecules that scientists would not have thought of.

“In terms of the increase in what we can look for, it’s practically limitless,” said Michael Webb, an assistant professor of chemical and biological engineering at Princeton and one of the study’s two lead authors. “Using machine learning to drive our search accelerates discovery by an amount that is difficult to quantify but very important. You may be spinning your wheels for a long time if you continue to rely on systematic search or testing.” -mistake”.

Led by Webb and Adam Gormley, an assistant professor of biomedical engineering at Rutgers, the researchers published their findings in the journal advanced materials.

In developing their system, the team turned to three proteins with distinctive properties, including a protein found in horseradish that is widely used in hospitals and water treatment plants.

“If we could solve the problem of these three, then we could theoretically extend the same procedure to all kinds of enzymes,” said Roshan Patel, a graduate student in Webb’s lab and one of the first authors of the new paper.

While proteins perform all sorts of amazing feats in nature, they tend to be picky about the conditions they work with. Changes in temperature or exposure to solvents can stop them dead in their tracks. To harden proteins for use outside their native environments, scientists often reinforce them with specialized support materials, such as rebar in concrete, making these brittle structures more robust. That’s a key step in enabling a vast body of biomedical, environmental, and other industrial technologies.

But finding the perfect match between a protein and its supporting molecule means optimizing an astronomical number of options. Conventional methods are slow and largely unsystematic (think trial and error), which means that most possible solutions are left unexplored.

With the horseradish protein, the researchers started by creating 500 different supportive molecules based on the traditional and intuitive approach. Each support had some potential to strengthen the protein against harsh industrial conditions, but the researchers didn’t know much more than that. They then tested each of the 500 molecules as a support, collecting actual data on their performance, while at the same time tasking the computer model with making predictions about what they would find. Comparing the predictions with the findings allowed them to improve the computer model through a process of positive reinforcement, called reinforcement learning.

With the newly trained computer model, the researchers expanded their search to more than half a million potential support molecules. Each molecule represented a different recipe assembled from thousands of ingredients in various configurations. They ran the data through the model four times, each time looking for two things: molecules that would outperform the rest of the field, or molecules that had some interesting quality that might make the algorithm even more sophisticated.

“In the fifth race,” said Webb, “we took off the handcuffs. We said, okay, give us the best 24 [molecules] you can find.”

Compared to the molecules they identified using gut-based methods, the new machine-assisted approach found supporting molecules that worked more than five times better for the horseradish protein. When working with lipase, a protein that breaks down body fat, the results were much more dramatic. The new system found a support molecule that improved performance by about 50 times compared to initial options, even pushing the protein function better outside of their native environment than in their natural state.

“There are many things you can manipulate about [these molecules]including the chemistry of their underlying units, their size, their architecture, their sequence,” Webb said. “All of those things can affect properties in a way that you could exploit” for a useful application.

Webb said they could streamline the process and speed it up even more by integrating the machine learning model with the physical robotics system in place. Much of the early work was done by sending data back and forth between the two labs.

He also pointed out specific applications the team was starting to work on, where to find molecules to stabilize proteins could lead to transformative solutions: a new way to recycle hard-to-break plastics and a non-invasive treatment for spinal cord injuries.

“There is an opportunity to track and figure out more precisely why these things work and the conditions under which they work,” Webb said.