Artificial Intelligence Calculates Anti-Aging Properties Of Compounds
By Deborah Borfitz
August 31, 2021 | Artificial intelligence (AI) has been paired with one of the simplest of organisms—the nematode Caenorhabditis elegans—to enlighten the scientific community about the physical and chemical properties of drug compounds with anti-aging effects, according to Brendan Howlin, reader in computational chemistry at the University of Surrey (U.K.). The predictive power of the methodology has just been demonstrated using an established database of small molecules found to extend life in model organisms.
The 1,738 compounds in the DrugAge database were broadly separated into flavonoids (e.g., from fruits and vegetables), fatty acids (e.g, omega-3 fatty acids), and those with a carbon-oxygen bond (e.g., alcohol)—all heavily tied to nutrition and lifestyle choices. Pharmaceuticals could be developed based on that nutraceutical knowledge, including the importance of the number of nitrogen atoms, says Howlin.
Unlike prior efforts using AI to identify compounds that slow the aging process, Howlin used machine learning to calculate the quantitative structure–activity relationship (QSAR) of molecules. The model utilized 20% of the DrugAge compounds for the test set to learn which chemical properties were important. The information was then used on the remaining 80% to train the model to identify compounds with those properties, he explains.
As described in a recently published article in Scientific Reports (DOI: 10.1038/s41598-021-93070-6), the study builds on the work of another researcher (Diogo Barardo, University of Liverpool) who a few years ago built a random forest model to predict whether a compound would increase the lifespan of C. elegans based on data in the DrugAge database. His top-30 list of predictive molecular features referred to atom and bond counts as well as topological and partial charge properties of the substances.
The nematode is frequently used in age-related research because it has many of the organ systems present in more complex animals and has a short lifespan of 20 days, says Howlin. That makes it possible to conduct experiments that are not practical in either mice or humans.
AI is now routinely employed by pharmaceutical companies in lieu of having hundreds of organic chemists testing “every possible variation of every possible compound” to see what works, says Howlin. In fact, AI is adding speed to virtually every stage of the drug discovery process by reducing repetitive, time-consuming tasks.
That breadth is represented by research underway at the University of Surrey, he continues, where AI-savvy scientists are helping to identify hits and leads, modify compounds to optimize their activity, predict how drugs are metabolized and affect the liver, and train the next generation of students in practical, real-world applications of machine learning algorithms.
Howlin has been actively involved in anti-aging drug design for many years now. He is one of the inventors of bi- and tri-aromatic compounds as NADPH oxidase 2 (Nox2) inhibitors, which are thought to have potential in treating a wide range of common, often age-related, diseases as well as aging itself.
NADPH oxidase is an enzyme made by the body to defend against bacterial infections, says Howlin. But if it doesn’t turn off like it should, it produces oxidative stress that can damage the blood vessels and trigger diseases of aging.
The AI-based prediction model was a “sideline” project to see if the research team could provide industry with some drug discovery clues. Employing the latest version of the DrugAge database, it expands the number of identified molecules with anti-aging properties to 395 from the 229 previously identified by Barardo, while the volume of compounds that did not increase lifespan held steady at 1,163, Howlin reports.
The study describes several compounds—the flavonoids rutin and hesperidin (the predominant phenolic compound in orange extracts) and the organooxygen compounds lactose and sucrose—which were previously found to be longevity-promoting in experiments on C. elegans. Future work will need to consider dosage, since it can impact whether a substance is beneficial or detrimental, he notes.
In addition to rutin (abundant in many plants), further in vivo testing may be warranted for gamolenic acid (plentiful in evening primrose oil and black currant oil), lactulose (shown to effectively treat chronic constipation in the elderly patients), and rifapentine (an antibiotic approved for the treatment of tuberculosis) based on the predictive exercise.
Moving forward, the machine learning model could be applied to any database to calculate the properties of different compounds, Howlin says. Many such databases are the property of pharmaceutical companies and could be tapped as a first step to improving human health by helping people age better.
University of Surrey researchers could also be supplementing their own aging research by finding new active compounds they can test alongside their experimental Nox2 inhibitors, he adds.