Automation speeds up the search for stable proteins

Automation speeds up the search for stable proteins

Researchers have settled on an important protein by harnessing the dual power of robotics and machine learning. They also showed that the system works with a range of proteins. Their robotics platform installed augmentation particles ten times faster than pioneering methods. And the machine-learning model opened up huge new possibilities for finding the right combination of materials, cutting months or even years of time for scientists to pair proteins with ideal struts. Image courtesy of researchers. Credit: Michael Webb/Princeton University

By harnessing the power of robotics and machine intelligence, researchers from Princeton Engineering and Rutgers University have found a way to design stable proteins in a fraction of the time at present state of the art. The team’s robotics platform is speeding things up more than tenfold, and their computational approach finds solutions anywhere from weeks to years faster than is possible by human intelligence alone.

Stabilization of proteins is a major challenge for research into drug creation, biofuel production, and plastic recycling. Currently, scientists use their knowledge of chemistry to estimate chemical compounds that will combine well with proteins under various conditions. The traditional approach uses trial and error to improve results. This painstaking method can take months as scientists create and test samples of molecules, and often it leads nowhere.

In the new system, engineers use machine learning Model To determine Chemical compounds Most likely to stabilize the desired proteins. The form helps narrow down hundreds of thousands of possibilities to a small number of potential candidates. A robotic assembly platform produces samples of molecules for evaluation. Combining the robotic platform and machine learning model Show results in less than a few days.

This twin-turbo approach offers an added advantage: due to its ability to pass huge amounts of data, the machine learning model often recommends candidate molecules that would otherwise not occur to scientists.

“In terms of the increase in what we can look for, it’s pretty much unlimited,” said Michael Webb, associate professor of chemical and biological engineering at Princeton University and one of the study’s senior authors. “Using machine learning to direct our research is speeding up discovery by an amount that is difficult to quantify but is very important. And you can spin your wheels for a very long time if you continue to rely on research or systematic trial-and-error.”

Led by Webb and Adam Gormley, associate 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 distinct properties, including a protein found in horseradish that is widely used in hospitals and water treatment plants.

“If we could solve the problem for these three, 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 research paper.

While proteins perform all kinds of amazing jobs in nature, they tend to be selective about their working conditions. Changes in temperature or exposure to solvents can stop it in its tracks. To strengthen proteins for use outside of their native environments, scientists often reinforce them with specialized supporting materials — such as steel rebar in concrete — which make these fragile structures even more robust. This is an essential step to enable a wide range of other biomedical, environmental and industrial technologies.

But finding the perfect match between a protein and its support molecule means optimizing an incredible number of options. Conventional approaches are slow and largely unsystematic – think trial and error – which means that most possible solutions will not be explored.

Using the horseradish protein, the researchers set out to make 500 different support molecules based on the intuitive, traditional approach. Each support had some potential to harden the protein against harsh industrial conditions, but researchers didn’t know much more than that. Then they tested each of the 500 molecules as a prop, gathering real data on its performance, while at the same time tasking a computer model to make predictions about what they would find. Comparing predictions to outcomes allowed them to improve the computer model through a process of positive reinforcement, called reinforcement learning.

Using their newly trained computer model, the researchers extended their search to more than half a million potential support molecules. Each molecule represents a different recipe grouped together from thousands of ingredients in different 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 more complex.

Webb said “In the fifth round, we took off the shackles. We said, OK, give us a top 24 [molecules] You can find.”

Compared to the molecules they identified using intuition-based methods, the new machine-assisted approach found support molecules that perform five times better for the horseradish protein. When working with lipase, a protein that breaks down fats in the body, the results were even more dramatic. The new system has found a support molecule that improves performance about 50 times compared to the initial options, and even pushes protein to work better outside of their original environment than if they were in their natural state.

“There are a lot of things you can manipulate [these molecules], including the basic units’ chemistry, size, structure, and sequence,” Webb said. All of these things can affect properties in a way that you might exploit” for a useful application.

Webb said they could simplify and speed up the process even more by merging a file machine learning model With an on-site physical robotics system. Much of the initial work was done by sending data back and forth between the two testers.

He also referred to specific applications that the team started working on, where they were found molecules Stabilizing proteins could lead to transformative solutions: a new way to recycle hard-to-break plastics and a non-surgical treatment for spinal cord injuries.

“There is an opportunity to do follow-up and find out more precisely why these things work and the conditions in which they work,” Webb said.

Fluorescent particles detected by quantum chemistry and machine learning

more information:
Matthew J. Tamasi et al., Machine learning on a robotic platform for polymer-protein hybrid design, advanced materials (2022). DOI: 10.1002 / adma.202201809

Introduction of
Princeton University

the quote: Automation Accelerates Search for Stable Proteins (2022, Aug. 2) Retrieved on Aug 2, 2022 from

This document is subject to copyright. Notwithstanding any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.