|
Getting your Trinity Audio player ready...
|
Researchers at the Pritzker School of Molecular Engineering (PME) at the University of Chicago have harnessed the power of machine learning to revolutionise vaccine design. In a study published in Chemical Science, the team utilised artificial intelligence (AI) to guide the discovery of small molecules called immunomodulators, potentially paving the way for more effective vaccines and robust immunotherapies for cancer treatment.
The challenge lies in navigating a vast chemical space where the number of drug-like small molecules has been estimated to be a 10^60—surpassing the number of stars in the visible universe. To tackle this complexity, the researchers employed machine learning to guide high-throughput experimental screening, providing a systematic and efficient approach to identify molecules that could induce the immune response.
Professor Aaron Esser-Kahn, emphasised, “We used artificial intelligence methods to guide a search of a huge chemical space. In doing so, we found molecules with record-level performance that no human would have suggested we try.”
The AI-guided approach marked a potential first in the field of vaccine design. Professor Andrew Ferguson, who led the machine learning efforts, highlighted the transferability of tools from drug design to immunomodulator discovery. While machine learning is commonly employed in drug design, its application in this manner for immunomodulators is an advancement.
Immunomodulators alter the signalling activity of innate immune pathways within the body, particularly the NF-κB and IRF pathways. These pathways are crucial in inflammation, immune activation, and antiviral response. Previous high-throughput screens identified molecules that enhanced antibody response and reduced inflammation when added to adjuvants in vaccines.
The team integrated the results with a library of nearly 140,000 commercially available small molecules to expand the pool of candidates further. Graduate student Yifeng (Oliver) Tang utilised active learning, a machine learning technique, to efficiently navigate experimental screening through molecular space. This iterative process, guided by the AI model, uncovered high-performing small molecules that had never been identified.
After four cycles and sampling only about 2% of the library, the team discovered molecules that improved NF-κB and IRF activity. One standout molecule demonstrated a three-fold enhancement of IFN-β production when delivered with a STING agonist, holding promise for more potent cancer treatments.
Professor Esser-Kahn highlighted, “The challenge with STING has been that you cannot get enough immune activity in the tumour or have off-target activity. The molecule we found outperformed the best-published molecules by 20%.”
The researchers identified several generalist immunomodulators capable of modifying pathways when co-delivered with agonists. These molecules could have broad applications across various vaccines, simplifying the path to market.
The team plans to continue this process, searching for more molecules and urging collaboration within the scientific community to share datasets for more exploration. Their future goals include screening molecules for specific immune activity, such as activating certain T-cells or discovering combinations that provide better control of the immune response.
In the quest to find molecules that can effectively treat diseases, the intersection of machine learning and immunomodulator discovery opens new possibilities for advancing medical science and developing innovative solutions for vaccine design and cancer immunotherapy.
