In recent years, AI technology has developed to incredible heights and its potential applications in the sciences have been exponentially increasing. From self-driving cars to unlocking the mysteries of dark matter, artificial intelligence has become a powerful tool of unprecedented breadth.
Now, scientists are exploring another possible use for AI: generating original proteins from scratch that can act as promising therapies for previously incurable medical conditions.
Through natural language processing (NLP) models and deep learning algorithms, researchers are able to design manufactured proteins with therapeutic functions far beyond what was ever thought possible before using traditional methods—allowing us to be one step closer to revolutionizing healthcare and disease treatment.
What Had Happened?
Researchers have made a remarkable breakthrough in the field of artificial intelligence – an AI system created to read and respond to English text has successfully synthesized enzymes from scratch. These synthetic proteins proved as effective in laboratory trials as their natural counterparts, showcasing that even machines can grasp some fundamental concepts of biology.
Created by Salesforce Research, ProGen applies next-token prediction algorithms when constructing amino acid sequences into functional proteins. This advancement brings us one step closer to advancing our understanding of language processing technology!
James Fraser, Ph.D., a professor of bioengineering and therapeutic sciences at the UCSF School of Pharmacy, recently released groundbreaking research that made headlines in Nature Biotechnology on January 26th. His work involved artificial enzymes with sequences only 18% similar to one another yet still possessing an activity equal to HEWL (a natural enzyme).
Even more remarkable was how these two proteins were 90-70% identical across all known varieties!
How precisely AI can generate human proteins?
In a remarkable trial, scientists unveiled the incredible productivity of ProGen-generated enzymes – their functions remain largely intact even with only 31.4% similarity to naturally occurring proteins! Such resilience is especially impressive when considering that just one mutation in an average protein can completely deactivate it.
The capacity for this technology lies far beyond what any human designer could ever achieve; there are 20300 possible combinations of amino acids within each lysozyme – more possibilities than all people who have ever lived combined and enough data points to develop every single grain on Earth or atom in space!
The model’s ability to effortlessly create functional enzymes out of the never-ending possibilities is simply astounding. If you are seeking their complete study, check there.