Biology, the study of life, is mainly the study of proteins. Unlike small chemical molecules, these proteins move, grow, shrink and interact. Folding and unfolding in a complex dance driven by code has so far troubled scientists to a large extent.
In December, DeepMind shocked the world Prove It can solve the problem 50 years ago, that is, predict how a protein folds based on its amino acid sequence.Just last month, this London-based company Announce Its “AlphaFold” almost solves the “protein folding problem” of all natural protein sequences that we constitute: 98.5% of the sequences in the human body.
Founded in 2010, acquired by GoogleIn 2014, Alphabet’s artificial intelligence subsidiary initially focused on simulating the neural network of the human brain, but later evolved to solve a more complex problem: for any given protein, unlock the most stable structure among many possible structures.Protein misfolding is At the root Many neurodegenerative diseases, such as Parkinson’s disease and Alzheimer’s disease.
The amino acid sequence of a protein determines its three-dimensional atomic structure, but there are at most one thousand such amino acids in each chain. Citizen scientists have tried crowdsourcing solutions in which players all over the world attacked the next level of Rubik’s Cube in one way. online game Called “FoldIt”, but for many more complex folds, it will take years before they find a solution. Since 1994, bioinformatics scientists have been collaborating to crack codes and pass key assessments of protein structure prediction (CASP).
By Win the biennial challenge of CASP 2020 DeepMind demonstrated the predictive power of Alphafold’s latest version, Alphafold2.With median Global distance test (GDT) scored 92.4 points out of a total of 100 points for all goals, beating the best of the other contestants that year by more than 20 points.
July 22, DeepMind shared According to the results of its research, the winning AlphaFold2 team used artificial intelligence to predict the structure of more than 20,000 human proteins, as well as the structures of almost all known proteins produced by 20 model organisms, such as Escherichia coli, Fruit flies, yeast, soybeans and Asian rice, there are about 365,000 predictions in total.
DeepMind’s achievements—— Compared by many people The mapping of the human genome-has an explosive impact on many health and life science industries: small molecule and biopharmaceutical development, diagnostic test development and even disease prediction.
Life science lawyers say it also brings new challenges with Provide opportunities for companies seeking patent protection for protein interaction discovery. Can your company profit from using these treasure troves of biological data?
Artificial intelligence platforms for drug discovery have just begun
Life Science Legal Expert Definition AlphaFold2’s underlying securities Platform technology as a “great change” from many Protein Predictive Modeling Tool It has existed before, including the original AlphaFold.
In a telephone interview, Kevin O’Connor, partner of Neal Gerber Eisenberg’s intellectual property business group, predicted that DeepMind and other artificial intelligence companies will only continue to use this platform tool to research, identify, treat, and cure diseases.In doing so, extensive patent protection will be sought to protect these two systems with The method they use to generate 3D structures has great advantages.
“The platform may even be relevant to small molecule drug discovery efforts, helping companies evaluate the binding of drug candidates to the proteins represented in their 3D structures,” O’Connor said.
We cannot apply for a patent for life, but we can apply for a patent near-Life
Although DeepMind disclosed the natural protein in its 3D structure dump in the Supreme Court in 2013 Decide exist countless In the following cases, any synthetically derived or unnaturally structured protein may be.
“If it is not a natural protein, thematic protection is not an obstacle, so any such discovery is eligible,” O’Connor said. “I can imagine traditional statements based on primary amino acids and based on cell structure.”
He explained that if DeepMind or elsewhere has not disclosed structural information, and it does not reflect the structure of the protein under natural conditions, the company may even be able to apply for a patent for the natural protein. He said that if the unnatural confirmation of the protein is attractive, then this structural statement may be a way to solve the problem.
“If you start with natural protein, in order to facilitate manufacturing or delivery or some other advantages, it is not natural confirmation, you avoid the subject qualification problem, because it is not a product in a natural state, so you are in the compound or franchise A layer of protection is added around the right,” he explained.
For example, O’Connor can imagine a future scenario where someone uses this AI modeling to discover new propositions for antibody therapy. If there are different folding processes or folding results based on different conditions—for example, by using different esters or salts to change the pH level—one can apply for a patent for one or more unpublished alternative confirmations.
Transfer research funds
Given the complexity of the protein folding problem and its potential relevance to medical breakthroughs, it is reasonable for universities and companies to spend a lot of research budgets trying to get it right.
“The completion of the DeepMind project and the global availability of its research results should bring new opportunities, which may indicate that priorities in the field of R&D and drug discovery are changing,” John Hoff, legal counsel of Certara, who focuses on model-informed drug development Man said about technology and service in a telephone interview.
Hoffman and O’Conner agree that this discovery may affect research and development (R&D) budgets, redistributing resources from folding problems to discovering new receptors for small molecules and biological therapies. “The race to win patents for these key receptors is ongoing,” Hoffman said.
As a neuroscientist with a background in pediatric inflammatory diseases, O’Connor pointed to a huge opportunity for innovation around the diagnosis of protein misfolding diseases, especially.
“What we have learned here can be used as a starting point for diagnosis and future treatment, which has not been before,” O’Connor said.
As the cornerstone of life, protein converts DNA instructions into intertwined amino acid bands, which are tightly packed when formed, forming strong, stable, and sometimes even beautiful structures, such as Lens of the human eyeHowever, if these crystal proteins disintegrate, it will cause cataracts to become cloudy.
O’Connor believes that the ability to compare the structure of healthy cells with those associated with misfolded diseases is very valuable in itself. This attribute provides further power for this artificial intelligence platform as it is leveraged through partnerships with research institutions and pharmaceutical companies at all stages of discovery and development. In fact, some people claim that AlphaFold2 proves that the role of artificial intelligence is “Meta-technology“Allows the science of finding a needle in a haystack that changes the course of history.
Turbocharging other technologies
Hoffman foresaw the patent challenges of DeepMind’s open framework, but also envisioned opportunities for his company.
“I think DeepMind may encourage the use of computer-simulated bio-simulation models (using precise, customized computer-based simulations instead of on-site subjects) in conducting trials to predict the best dosing regimen and identify potential additional bio-simulation use cases ,” Hoffman said.
Part of the research ecosystem and evolving understanding
Paul Workman, Chief Executive Officer, Institute of Cancer Research (ICR) praise DeepMind opened up access to AlphaFold2 and pointed out its role in stimulating further AI research. He mentioned the work of scholars who recently published their academic papers. result It was shown that they used a “three-track neural network” called RoseTTAFold, which obtained almost as accurate structural predictions as DeepMind.
“In general, I believe AlphaFold2 is a major advancement in the journey of predicting the 3D structure of life proteins. It will have a profound impact on accelerating our overall understanding of the basic structure and function of life and disease,” Walkman wrote“The journey continues.”
Photo: AVNphotolab, Getty Images
