When COVID-19 emerged as a global catastrophe, the scientists had only one goal: to come up with the possible treatments for the virus SARS-CoV-2, which was quickly changing its form. Now, scientists appear to be harnessing the potential of artificial intelligence (AI) to explore unexplored territories in science. One such approach introduced by Stanford University and Chan Zuckerberg Biohub researchers is the Virtual Lab, a framework where nanobodies—the tiniest of the versatile versions of antibodies that bind and neutralize the virus regardless of its variants—have been designed by Artificial Intelligence with support from a cross-functional group of scientists.
Virtual Lab demonstrates that AI has the potential to not only assist in performing science but also turn it into an active participant in a new age of collaborative networks with humans. So, what is the Virtual Lab precisely, how does it function, and why is it such a revolutionary development? Let’s break it down.
The Big Idea Behind the Virtual Lab
The Virtual Lab, to put it simply, is a unit of LLMs that imitates real managers and scientists. Simply put, it’s like a think tank that is created and operated in the virtual world. Each AI agent has a specific role: a Principal Investigator (PI) to guide discussions, a Machine Learning Specialist, an Immunologist, a Computational Biologist, and even a Scientific Critic to challenge ideas and improve outcomes.
These agents conduct meetings that are similar to the standard meetings of scientists to reach conclusions on certain matters, resolve issues, or even program certain applications. These meetings are attended by an expert researcher who performs the role of AI synthesis and animates the research work. The singular feature of Virtual Lab is the manner through which it enables the cross-disciplinary approach to work—thus enhancing intricacy in processes while achieving a smooth and robust approach to solutions.
The team decided to develop nanobodies able to bind with SARS-CoV-2 nanobody therapeutics. Nanobodies are particularly interesting because they are very small, easy to engineer, and very potent in targeting viral proteins.
How the Virtual Lab Tackles Science
The operation of the Virtual Lab encompasses several phases, with the order of these phases resembling how research teams operate when conducting actual studies. Let’s go through the sequence of events that took place in the workflow related to the nanobody design:
Project Planning: The agents first deliberated about whether to create new nanobodies from scratch or to come up with variants of the currently existing ones. There was overwhelming consensus among agents, who spoke from varying perspectives. For the Immunologist, the goal was to modify existing nanobodies to expedite the process. The Scientific Critic, on the other hand, was more interested in the newer structures.
Tool Selection: Some of the agents made a bid and a choice as to the computational tools that were required. For instance, AlphaFold was selected to predict the structure of proteins and Rosetta to model interactions between proteins. While a Machine Learning Specialist set the learning algorithms, a Computational Biologist made sure the tools were appropriate for the nanobody analysis.
Code Development and Refinement: For instance, the agents used Python scripts to evaluate the sequences of nanobodies in one phase. Important modifications were made after some of the Scientific Critic’s observations showed that there were shortcomings ignored during the drafting stages of the first code versions. Observations that were addressed improved the function and precision of the various modifications. Such iterative function feedback prevented the final workflow from being unstable and ineffective.
Parallel Problem-Solving: To increase productivity, the Virtual Lab conducted several parallel meetings, with different sets of agents addressing the same question and attempting to find the same solution without consulting each other. Their outputs were later merged, ensuring the best ideas made it into the final pipeline.
The Results: A Breakthrough in Nanobody Design
At the Virtual Lab, delivering results, as opposed to just designing nanobody candidates, was possible. Due to AI’s assistance, the team was able to create a pipeline that resulted in the generation of 92 candidates, which were experimentally shown to bind to several different SARS-CoV-2 viruses.
In particular, Nb21 was a successful candidate possessing more effective binding in infected JN.1 and KP.3 viral variants. This achievement emphasizes the competence of the Virtual Lab, which is capable of scanning through trillions of nanobody mutations and finding the most likely ones faster than usual processes.
To keep making these advancements, the meetings were quite short, approximately 5-10 minutes in duration. The human researcher’s contribution was low, about 1.3% of the total writing effort. On the other hand, over 120,000 words were produced by the AI agents as research notes, Python-based scripts, and project summaries, among others.
Strengths of the Virtual Lab
The Virtual Laboratory has certain advantages in dealing with problems, as it has a diverse approach to the issues at hand.
Interdisciplinary Collaboration: Everybody involved in the discussion has a different contribution to make, and hence, the attended discussion is very lively, like how people work in real-life teams. For example, the Computational Biologist was responsible for the molecular docking, and the Machine Learning Specialist implemented algorithms for the evaluation of the Nanobody.
Efficiency: There are just certain portions that repeat and can have their code written and/or the data analyzed, which can even be turned into a Virtual Lab, meaning wider opportunities for professionals in the human industry.
Feedback-Driven Improvement: Scientific Critic feedback implementation is crucial and guarantees all outputs are improved and not just facts noticed. In fact, it has been even reported that similar quality results have been reported frequently.
Challenges and Future Potential
However, despite all its advantages and the fact that the Virtual Lab is a breakthrough, it does pose problems:
Outdated Knowledge: The AI agents are developed using already existing data, so they might not be up to date with the new developments. For example, the agents once suggested such tools as AlphaFold and required guidance from the human investigator to change them.
Prompt Engineering: This details the process that the agent goes through in order to be effective. Most of the time, the researcher had to make changes to these parts to get the best out of the virtual laboratory.
Despite these challenges, the Virtual Lab’s modular design means it can evolve with advancements in AI and scientific tools. Future versions could integrate real-time data retrieval and more advanced LLMs, making it even more powerful.
Conclusion
The achievement of the Virtual Lab in the designing of the SARS-CoV-2 nanobody candidates is indeed a game-changing approach to achieving scientific goals. It is not a simple case of improving speed in workflows but rather how AI as a science is being resolved. With further developments in AI, this Virtual Lab could be used in laboratories across the globe to help scientists address critical challenges in an incredibly efficient and accurate manner. In the meantime, it is a proof of concept of what humans and machines, if aligned, can achieve together.
The Virtual Lab is the future, that is, interactive, cross-disciplinary, and driven by AI.
Article Source: Reference Paper | Code for the Virtual Lab is available on GitHub.
Disclaimer:
The research discussed in this article was conducted and published by the authors of the referenced paper. CBIRT has no involvement in the research itself. This article is intended solely to raise awareness about recent developments and does not claim authorship or endorsement of the research.
Important Note: bioRxiv releases preprints that have not yet undergone peer review. As a result, it is important to note that these papers should not be considered conclusive evidence, nor should they be used to direct clinical practice or influence health-related behavior. It is also important to understand that the information presented in these papers is not yet considered established or confirmed.
Follow Us!
Learn More:
Anchal is a consulting scientific writing intern at CBIRT with a passion for bioinformatics and its miracles. She is pursuing an MTech in Bioinformatics from Delhi Technological University, Delhi. Through engaging prose, she invites readers to explore the captivating world of bioinformatics, showcasing its groundbreaking contributions to understanding the mysteries of life. Besides science, she enjoys reading and painting.
