Gero Taps Quantum Computing and AI To Tackle Diseases Of Aging
By Deborah Borfitz
August 22, 2023 | A movement is afoot to apply quantum computing to the puzzling business of designing drugs with the right molecular properties. The number of possible drug-like molecules is massive and most of them have never been synthesized, but quantum computers are the perfect fix for navigating this uncharted chemical space since molecules are quantum objects whose madcap behavior follows fundamental laws of physics, according to Peter Fedichev, cofounder and CEO of Singapore-based Gero.
In May, Gero (Scientific Reports, DOI: 10.1038/s41598-023-32703-4) as well as Insilico Medicine (Journal of Chemical Information and Modeling, DOI: 10.1021/acs.jcim.3c00562) published papers pointing to the drug discovery potential of quantum computing devices. A month later, SandboxAQ (an Alphabet company), announced the launch of its new drug-discovery AQBioSim division that will use artificial intelligence (AI) and quantum technologies for in silico simulation of molecular interactions.
Fedichev describes Gero as “experts in modeling human health and applying AI to find the signatures of aging and diseases.” It is not new to computer-aided drug discovery, he says, referencing the company’s most recent collaboration with New York University focused on finding molecules that may re-sensitize drug-resistant bacteria to known antibiotics (Science, DOI: 10.1126/science.abd8377).
“I happen to have a physics background, and this explains why I am so curious about exploring what quantum computers may offer for drug discovery,” says Fedichev. They promise to help solve problems that are “almost inaccessible for classical computers.”
Currently, available quantum devices are quite small and used primarily for research purposes, he notes. The challenge is that qubits, the basic building blocks of quantum computers, are notoriously fragile. Quantum information tends to degrade or get lost as a calculation progresses unless it’s protected from the ambient “noise” such as temperature fluctuations, electromagnetic radiation, or even cosmic rays.
Notable progress on this front comes from Google, where scientists recently reinstated their claim of “quantum supremacy,” says Fedichev. In a study published on the pre-press server ArXiv, they completed a computational task on a quantum computer that would have taken a classical supercomputer 47 years to finish. Separately, in an article that appeared in Nature (DOI: 10.1038/s41586-023-05954-4) in May, Google reports that it demonstrated building blocks of a topological quantum computer—opening the door to “fault-tolerant” quantum computation.
“With the rapid advances in quantum computing, it's not a question of if but when these powerful machines will be large enough to tackle practical problems,” says Fedichev. In the Scientific Reports paper, Gero researchers showcased the potential of a prototype quantum algorithm running on a commercially available computing device as a testbed for drug discovery by generating over 2,000 novel chemical structures with medicinal chemistry and synthetic accessibility properties in the ranges typical for molecules from the ChEMBL dataset.
“We built a generative model that can be used for generating examples of novel molecules, much like Midjorney generates unique images,” he says. “Generative models work by learning from a multitude of examples—in our case, a vast library of biologically active molecules.
“Generative models analyze the examples [molecules or images] they're given, but they don't simply memorize them,” Fedichev continues. “Instead, they use this analysis to create entirely new, yet realistic, molecules or images.” In the case of molecules, all the models need are enough training examples of drug-like and biologically active compounds to produce novel molecules that both resemble these compounds and possess the desired properties—“including the potential for actual synthesis.”
Being quantum objects, drug-like molecules possess many properties (including their synthetic availability) that are hard to predict without the aid of quantum computing. Gero has demonstrated a hybrid quantum and classical machine-learning system that is both small enough in its quantum layer to fit the size restrictions of currently available quantum devices and rich enough to learn representations of realistic drug-like molecules, says Fedichev.
The system that generated the novel chemical structures in the study was set up on a D-wave quantum device. The proposed molecules, in addition to being synthetically accessible in terms of structure, for the most part had the properties characteristic of drug-like molecules. Most importantly, “the overwhelming majority of the suggested molecules turned out to be also novel,” Fedichev says. “Less than 1% of them are similar to a previously published or otherwise listed molecule from research databases.”
Gero’s hybrid quantum-classical model, in the form of a Restricted Boltzmann Machine, is a steppingstone toward fully quantum models, says Fedichev. “Classic generative models rely on random-number generators to produce representations leading to novel samples. Quantum systems may exhibit far richer levels of uncertainty and correlations and therefore it is a good idea to let another quantum system, such as a quantum computing device, generate inputs leading to novel samples.”
The idea here is to use a quantum system (Quantum Boltzmann Machine running on a quantum computing device) to produce representations that are subsequently “unpacked” by the classic neural net into human-readable representations of molecules, he explains. “The outputs of such systems are classic and could be combined with other neural networks, helping to optimize other parameters, such as their ability to bind specific targets.”
Realities of Aging
Gero’s mission is to stop human aging by applying modern machine learning tools to unravel the dynamics of senescence and chronic diseases, says Fedichev. “We partner with experts in chronic diseases to understand the effects of aging on the disease progression and help identify targets for novel drugs backed by real-world medical data.
The company is building models of human health from large biomedical datasets and applying them to understand aging and chronic diseases. Such models can provide markers of diseases, in terms of risks or disease progression, he says. GeroSense, which uses wearable sensor data and artificial intelligence for predicting life expectancy, is one example of the approach. But it is in overlaying the models with genetics or other molecular data that Gero’s algorithms have the power to yield targets for therapeutics.
Fedichev’s theory is that aging is hard to reverse because it is the result of the cumulative effect of countless unrelated processes that could go wrong. The body’s ability to recover from these various stresses gradually weakens. His money is on tapping technology to help slow the pace of aging, as has in fact been accomplished in nature by creatures like the naked mole rat that defy the biological law of aging and have remarkable resistance to age-related diseases.
In a research collaboration with Pfizer announced in January, Gero’s machine learning technology platform will be used to discover potential therapeutic targets affecting reversible manifestations of fibrotic diseases using large-scale human-based data. The deal is not concerned with quantum drug discovery systems, says Gero cofounder Maxim Kholin.
Times are Changing
Only a few years ago, quantum computers couldn’t do anything practical, says Kholin. “Many people still believe quantum computers are exotic machines that are far from being useful.” But times are changing to the point of talking quantum supremacy in selected tasks, he notes. Those tasks could well include drug discovery in another few years.
Modern AI and machine learning systems are “huge force multipliers in approaching complex problems,” Kholin says. Building hybrid quantum and classical algorithms, as Gero has done, could be a great way for AI and drug discovery experts to expose themselves to quantum computing.
As for human aging, “[it] cannot be completely reversed but we firmly believe that it can be effectively halted,” he adds. Gero’s research and discoveries have given its scientists a strong understanding of the irreversibility of aging that could serve as a guide in identifying high-quality targets against most age-related diseases—and perhaps provide more realistic expectations about what will radically extend life than is reflected in current popular trends such as cellular rejuvenation.