Artificial intelligence (AI) in medicine is becoming mainstream. As PwC points out, by harnessing the learning potential of AI, new technologies will enable early-stage disease detection and detailed diagnosis, giving health care workers better, more informed insights. It also helps you make decisions based on
But this is just the beginning. A new effort in AI has produced tools that can design human proteins that meet or exceed the performance of evolutionary-generated proteins.
Speed Up: Fundamentals of AI Protein Design
Using technology to aid in protein design is an established practice. For example, the online protein folding game Foldit harnessed the collective power of human intelligence to explore the structure of enzymes. In 2011, scientists tasked his Foldit player with finding the structure of a specific enzyme involved in HIV reproduction. In three weeks, an online player cracked the code, solving a problem researchers had been trying to solve for years.
Meanwhile, a new human protein program is using next-generation AI to both replicate structures found in nature and think outside the biological box. As ScienceDaily points out, AI acceleration began with researchers at the University of California, San Francisco, who fed the amino acid sequences of 280 million proteins into the machine learning (ML) algorithms that underpin AI tools. They then gave them time to think about what they had learned and begin to understand the patterns and connections that make these proteins work.
The team then provided AI with 56,000 lysozyme family sequences and context on how these sequences formed. Using this information, the tool generated 1 million potential protein sequences. 100 were selected for testing, 5 of which were engineered into engineered proteins and used in cells.
The results were impressive. Two of the proteins produced were able to break down bacterial cell walls in a manner similar to the naturally occurring hen egg white lysozyme (HEWL). Interestingly, although these two proteins performed similarly, only 18% of the structures overlapped. Furthermore, the scientists found that the AI-generated options still showed some effectiveness even when only 31.4% of the structures matched known proteins.
From prediction to production
Predicting the shape of existing proteins based on pattern recognition is another thing, but what about creating entirely new proteins? According to the American Association for the Advancement of Science, there are two broad approaches to this goal. I have. It is restorative and restrained hallucinations.
Inpainting uses an AI solution to fill in the blanks around the central feature. For example, an AI tool might be given a portion of a protein that binds well to a particular antibody. Equipped with only this information and knowledge of protein patterns, AI can pinpoint missing pieces and slowly build structures around the central protein.
On the other hand, suppressed hallucinations drive AI out of control. Instead of proving a central feature of the protein, the tool is only given a purpose, such as binding to carbon dioxide. AI solutions generate new proteins based on an understanding of their building blocks and interactions. Once the full hypothetical protein is imagined, the tool evaluates the potential validity of the results. Then keep what works and change what doesn’t, each time moving closer to your goal.
In other words, this is evolution by AI steroids. Instead of waiting for the efficacy of these proteins to be proven in a physical laboratory in their natural environment, AI tools can check millions of potential designs in days or weeks.
However, this is not a scientific silver bullet. AI-generated proteins hold great promise, but rather than exist in isolation, they are part of a larger biological system that can trigger unanticipated reactions that alter the way proteins work. I have.
What’s Next for Human Protein Design?
If all goes according to plan, the next step would be to use AI in healthcare to develop new treatments for humans, and even in vaccine development.
For example, according to HealthITAnalytics, a Harvard and Washington University School of Medicine study using both restorative and psychedelic techniques showed promising results. In one case, the tool created a new protein that can bind to the anticancer receptor PD-1. Another study showed that the protein could form the basis of a vaccine for respiratory syncytial virus (RSV), which can be fatal to vulnerable groups.
However, don’t expect designer protein pharmaceuticals to appear anytime soon. AI can hallucinate and create and test proteins, but scientists cannot escape the rigor necessary to assess the risks of these potential pharmaceuticals. In practice, this means more laboratory testing, controlled testing with animal analogues, and limited human testing before new treatments become available.
Despite the potential distance between the design and delivery of AI-generated proteins, this proof of concept makes a strong case. Given data and time, AI can not only mimic, but in some cases even evolve, to build something never seen before.
Interested in science and innovation? So are we. Learn more about our employees and Life at Northrop Grummanor check us out career opportunities See how you can participate in defining possibilities.
