Researchers at the University of Chicago Pritzker School of Molecular Engineering leveraged equipment learning to facilitate the discovery of novel substances that enhance defensive pathways. They identified a specific small protein with the potential to surpass existing immunomodulators on the market, marking a significant advancement in vaccination design. Jason Smith from the University of Chicago was involved in this breakthrough.
Immunomodulators, characterized as minute substances, hold promise for bolstering treatments and developing more potent cancer vaccines.
The sheer volume of approximately 10^60 drug-like molecules poses a challenge in pinpointing the compounds that elicit the desired immune response, far exceeding the number of recognizable celebrities.
Through the utilization of machine learning, the researchers pinpointed new defensive pathway-enhancing substances, highlighting one specific small molecule that shows promise in outperforming current immunomodulators available in the market. This groundbreaking discovery is detailed in the Chemical Science blog.
Professor Aaron Esser-Kahn, a co-author overseeing the experiments, emphasized the application of artificial intelligence techniques in navigating an extensive chemical landscape. He expressed enthusiasm about unearthing particles with unparalleled performance levels that traditional approaches might not have explored. The framework of this innovative process is set to be shared with the scientific community.
Furthermore, Professor Andrew Ferguson, the project lead for the machine learning initiative, noted the significant role of machine learning in drug design but highlighted its underutilization in immunomodulator discovery. The successful transfer of technology across different domains underscores the versatility and potential of machine learning applications.
Molecular Screening Utilizing Machine Learning
Immunomodulators operate by modulating the signaling activity of the body’s innate immune pathways. The IRF pathway is pivotal in antimicrobial response, while the NF-B pathway contributes to inflammation and defense stimulation.
Earlier this year, the PME team scrutinized 40,000 distinct molecule combinations to assess their impact on these pathways. By incorporating these molecules into adjuvants that enhance immune responses in vaccines, they observed heightened antibody responses and reduced inflammation with the most promising candidates.
Leveraging these initial findings alongside a library of approximately 140,000 commercially available small molecules, the team embarked on an iterative computational and experimental journey to uncover additional candidates.
Graduate student Yifeng (Oliver) Tang employed active learning, a machine learning technique blending exploration and exploitation, to efficiently navigate the experimental screening process through molecular space. This method not only identified high-performing molecules for experimental testing but also pinpointed underexplored regions harboring potential candidates.
The iterative process involved the model identifying promising candidates or areas necessitating further exploration, prompting the team to conduct high-throughput screening on those molecules before feeding back the insights into the active learning algorithms.
Graduate student Yifeng (Oliver) Tang, alongside Professor Aaron Esser-Kahn, graduate student Jeremiah Kim, and Professor Andrew Ferguson, reviewing the outcomes of the high-throughput empirical screen. Jason Smith from the University of Chicago
Enhanced Molecule Performance
After four phases and examining only about 2% of the library, the team unearthed high-performing small molecules that had not been previously identified. These top candidates exhibited an 83% increase in IRF activity, a 128% suppression of NF-B activity, and a 110% enhancement in the same.
One molecule, when combined with a Bite (modulator of interferon genes) agonist, triggered a threefold increase in IFN- output. STING receptors, known for bolstering immune responses in tumors, present a promising avenue for cancer treatment.
Esser-Kahn highlighted the newfound protein’s superior performance compared to existing substances, particularly in addressing the challenge of immune activity within tumors without off-target effects.
Additionally, the team identified “generalist” immunomodulators capable of modifying pathways when paired with agonists, potentially expanding their application in vaccines.
Ferguson underscored the versatility of these groups across various vaccinations, hinting at smoother market integration. The prospect of a single protein fulfilling multiple roles signifies a significant advancement in the field.
By discerning common chemical traits among the substances that induced desired behaviors, the researchers gained insights into the molecular structures identified through machine learning. This knowledge can guide the focus towards molecules exhibiting these properties or aid in the strategic engineering of novel compounds.
The team is poised to continue this exploration, welcoming collaborations to bolster their dataset and enhance their success rate. Their future endeavors aim to fine-tune immune responses through precise molecule combinations or tailored substances targeting specific immune activities like T cell activation.
Esser-Kahn expressed the team’s ultimate goal of identifying molecules with therapeutic potential to combat diseases effectively.
Through the strategic application of machine learning in guiding high-throughput experimental screening, the University of Chicago’s Pritzker School of Molecular Engineering addressed this critical challenge.
The study titled “Data-driven identification of innate immunomodulators using machine learning-guided high throughput screening” by Yifeng Tang et al. is published in Chemical Science (2023). DOI: 10.1039/D3SC03613H
