TL;DR:
- Generative AI has the potential to revolutionize antibody development and design.
- Antibodies play a crucial role in the immune system’s defense against infections.
- Researchers are using generative AI tools, such as protein language models, to suggest mutations for antibodies.
- AI-guided alterations have improved the effectiveness of antibodies against SARS-CoV-2, ebolavirus, and influenza.
- The AI-guided changes are unlikely to restore the effectiveness of COVID-19 antibodies against the Omicron variant.
- The suggested modifications to antibodies often occur outside the regions that interact with their targets.
- Generative AI has the potential to improve existing antibodies and create entirely new ones.
- AI could help in developing drugs for challenging molecular targets, including G-protein-coupled receptors.
- Antibody drugs could be designed to bind to multiple targets simultaneously, enhancing their therapeutic potential.
- Designing completely new antibodies is challenging due to the complex interactions within their structure.
- Progress has been made in creating new antibodies with AI, but more data on antibody-target interactions is needed.
Main AI News:
During the peak of the global pandemic, scientists raced against time to develop effective treatments for COVID-19. One promising avenue of research involved utilizing antibodies extracted from the blood of individuals who had successfully recovered from the disease. However, a groundbreaking study published in Nature Biotechnology has now revealed that generative artificial intelligence (AI) has the potential to revolutionize this labor-intensive process. By leveraging neural networks similar to those powering the ChatGPT AI platform, researchers are exploring the application of AI in antibody design.
Antibody-based drugs are a lucrative market, with annual worldwide sales exceeding US$100 billion. The integration of generative AI, which can generate text, images, and other content based on learned patterns, has the potential to accelerate drug development and unlock the potential of antibody therapies for diseases that have thus far eluded conventional design approaches.
The co-lead author of the Nature Biotechnology paper, Peter Kim, explains the immense interest in antibody discovery and engineering. The quest to enhance antibody properties and effectiveness has been a major focus within the scientific community. Stanford University biochemist Peter Kim, together with computational biologist Brian Hie, spearheaded this pioneering study.
Antibodies play a pivotal role in the immune system’s arsenal against infections. Their unique properties make them highly desirable in the biotechnology industry, as they can be engineered to target a wide range of proteins and manipulate their functions. However, developing antibodies with desirable traits and improving upon existing ones involves extensive and laborious screening processes.
To streamline these processes, Hie, Kim and their team turned to generative AI tools. They employed protein language models, a type of neural network trained on tens of millions of protein sequences, to explore the potential of AI in suggesting mutations for antibodies. While traditional language models like ChatGPT are fed vast amounts of text, protein language models are trained on protein sequences.
Previous studies have successfully utilized these models to design entirely novel proteins and accurately predict protein structures. His team utilized a protein language model developed by researchers at Meta AI, a division of the New York City-based tech giant Meta.
The model provided a handful of mutations for antibodies, even though it was trained on only a few thousand antibody sequences out of the vast pool of approximately 100 million protein sequences it learned from. Surprisingly, a significant number of the model’s suggestions improved the binding affinity of antibodies against SARS-CoV-2, ebolavirus, and influenza.
These AI-guided alterations resulted in enhanced capabilities of an approved Ebola therapy and a COVID-19 treatment, enabling them to more effectively recognize and neutralize the proteins used by the respective viruses to infect cells. However, it’s important to note that the AI-guided changes are unlikely to restore the effectiveness of the COVID-19 antibody against the Omicron variant and its subvariants.
Remarkably, many of the suggested modifications to the antibodies occurred outside the regions of the protein that typically interact with their targets. This discovery has captivated researchers, as the model appears to tap into information that even antibody engineering experts find non-obvious. Kim describes this revelation as a “holy cow, what’s going on here?” moment, signifying the extraordinary potential of AI in antibody design and engineering.
According to Charlotte Deane, an immuno-informatics researcher at the University of Oxford, generative AI has the potential to be a valuable tool for improving antibodies. She finds it fascinating and believes researchers will use this technology to enhance their existing antibodies.
However, many scientists are hopeful that generative AI can go beyond improvement and actually create entirely new antibodies that can bind to specific targets. This capability would be particularly beneficial for developing drugs targeting molecular entities that have proven resistant to other antibody-design methods.
Surge Biswas, the co-founder of Nabla Bio, a company in Boston, Massachusetts, working on this challenge, explains that generative AI could aid in addressing G-protein-coupled receptors, a family of proteins embedded in cell membranes that play a role in neurological disorders, heart disease, and various other conditions. Furthermore, generative AI could facilitate the design of antibody drugs capable of binding to multiple targets simultaneously, such as a tumor protein and an immune cell that can eliminate the tumor.
Bioengineer Possu Huang from Stanford acknowledges the power of protein language models in optimizing existing proteins, including antibodies. However, models trained solely on protein sequences might face difficulties in generating entirely new antibodies with the ability to recognize specific proteins. Nonetheless, progress is being made in this area.
In March, scientists at Absci, a biotech firm in Vancouver, Washington, reported a preliminary step towards creating new antibodies with the help of AI. Their model incorporated protein sequences and experimental data to generate new designs for crucial regions of an antibody-drug used in breast cancer treatment.
Designing completely new antibodies presents a challenge because their ability to recognize specific targets relies on flexible loops within the antibody structure. These interactions have proven challenging to model accurately using AI techniques. Huang’s team developed a generative AI tool last year that can create proteins capable of strongly binding to specified targets, such as snake venoms, by considering these loops. Applying a similar approach to antibody design might require more extensive data on how antibodies interact with their targets, which is currently limited.
Conlcusion:
The integration of generative artificial intelligence (AI) in antibody development holds significant implications for the market. The potential of AI to improve existing antibodies and create entirely new ones opens up new avenues for therapeutic advancements. This technology has the capacity to accelerate drug development, enhance antibody properties, and address molecular targets that have proven resistant to traditional design approaches.
The market for antibody-based drugs, already valued at over US$100 billion annually, stands to benefit from the increased efficiency and effectiveness brought about by generative AI. As AI continues to make strides in antibody design and engineering, it will undoubtedly shape the future of the biotechnology industry, unlocking novel treatments and transforming the landscape of healthcare.
