Quantum computing makes waves in drug discovery

Many of us have been mesmerized by the simple elegance of waves flowing on the water’s surface. Whether at a beach, on a lake, or watching the rain fall into a pond, seeing the ripples intersect, interfere, and separate can feel meditative. These simple waves, both peaceful and powerful, govern the universe at the quantum scale (think even smaller than atoms). By harnessing the behavior of these waves at the quantum level, scientists are gaining a deeper understanding of molecules and proteins, which can significantly accelerate drug discovery.

The drug discovery process is an expensive and time-consuming endeavor, often costing billions of dollars and taking decades for a single drug to reach the market. However, these molecules have the potential to save and extend lives, galvanizing researchers, companies, and governments to seek ways to address the speed and expense of finding new therapeutics. 

Machine learning is a computational technique that allows computer algorithms to learn from previous data and make predictions and decisions or to generate new data. This technology, often using generative artificial intelligence, is the foundation for programs such as ChatGPT. In drug discovery, machine learning can significantly speed up this process by analyzing vast amounts of data to identify and generate new potential drug candidates. 

Recently, researchers at St. Jude and the University of Toronto showed that quantum computing, a technology that exploits quantum effects (such as superposition, entanglement, and interference), could boost machine learning-based drug discovery to find better molecules faster, including for previously “undruggable” targets. The research, published in Nature Biotechnology, sets the stage for a future that leverages quantum research.

“This is the first time quantum computing has been successfully used for a drug discovery project that includes experimental validation,” said co-corresponding author Christoph Gorgulla, PhD, Center of Excellence for Data-Driven Discovery, Department of Structural Biology. “By augmenting our machine learning model with quantum computing, we outperformed similar, purely classical [computing] machine learning models in identifying promising therapeutic compounds.”

Scientists targeted the KRAS (Kirsten rat sarcoma virus oncogene homolog) protein to showcase quantum computing's potential for drug discovery. KRAS is one of the most mutated genes in cancers and is known to be a difficult target, often being called “undruggable.” The researchers used a classical computer to input a database of all molecules experimentally confirmed to bind to KRAS and trained a machine-learning model with this data. They also included over 100,000 theoretical KRAS binders obtained from an ultra-large virtual screen. 

After running the classical model, they fed the results into a filter/reward function that evaluated the quality of the generated molecules, allowing only those of sufficient quality to pass the filter. Afterward, they trained a quantum machine-learning model and combined it with the classical model to improve the quality of the generated molecules. They then cycled back and forth between training the classical and quantum models to optimize them in concert. Once completed, the models generated multiple novel molecules, called ligands, which were predicted to bind KRAS.

“We identified ligands for one of the most important cancer drug targets,” Gorgulla said. “We then validated our discoveries through experiments, finding two molecules with real-world potential for future evaluation. This study serves as proof-of-principle that quantum computing has the potential to enhance drug discovery greatly in the future.”

Quantum computing overcomes the limits of chemical calculation with waves

Chemistry is, by its nature, the interaction of molecules at the subatomic level (in particular, the electron level). These interactions are best described by quantum mechanics, the branch of physics concerned with the strange and complex world of the very small. Quantum computers are designed based on these same principles, making them better suited for analyzing the interactions of molecules over classical computers.

“Classical computers struggle to efficiently handle the thermodynamical and quantum mechanical calculations required to simulate a protein and all molecules that could potentially bind to it,” Gorgulla said. “But quantum computers can, in theory, be exponentially faster for these calculations, as they are built using these same principles and are therefore better equipped to simulate molecular systems on the quantum mechanical level.” 

While classical computers can, in principle, perform these simulations, the computational cost quickly becomes impractical — especially when using highly accurate quantum mechanical methods, which can be crucial for modeling molecular interactions with high precision.

“Classical computing uses bits, which are either zero or one, like a light switch, to perform operations in a stepwise arithmetic process. It’s like adding one plus one to get two, then adding another one to get three. While this might seem quick for simple calculations, it becomes very time-consuming as the calculations become more complex,” Gorgulla explained. “Quantum computing is different — it uses qubits. Qubits can be zero, one, or even zero and one at the same time. Instead of a switch, qubits can be pictured as a sphere like Earth. The north and south poles corresponded to zero and one, but the qubit’s value could be anywhere on the surface, effectively in an infinite number of positions, allowing the computer to explore multiple solutions simultaneously.”

The overlapping state of being both one and zero is called superposition. So, instead of a specific value, a qubit represents multiple states between one and zero. The real power of quantum computing comes from using qubits in superposition that are also “entangled.” Entangled qubits act together in a coordinated way, losing their individual identities and always acting and influencing each other. 

Unlike classical computers that perform arithmetic operations, entangled qubits perform operations that explore computations as waves. Just like waves on the ocean’s surface, these waves can interfere with each other, getting bigger (positive interference) or smaller (negative interference). The resulting waves represent a range of potential solutions to a problem rather than just one answer, all from a single run. The concepts of superposition, entanglement, and interference are used in the quantum simulation approach mentioned earlier, but also in other types of algorithms, such as quantum machine learning algorithms. 

In the study’s drug discovery pipeline, the concepts of entanglement and interference were crucial for improving the accuracy of their machine learning model in predicting whether compounds could bind to the chosen target, in this case, KRAS. “We found a few potential molecules for the most common KRAS mutants for which there are currently no drugs on the market,” Gorgulla said. “But we hope this is just the first of the many targets for which we will discover and validate viable lead compounds using this technology.”

Bringing quantum computing’s waves to St. Jude 

Gorgulla was recently appointed as an Assistant Member in the Center of Excellence for Data-Driven Discovery, Department of Structural Biology. One of his objectives is to develop new computational methods in his research group and the other is to help co-lead the institution’s use of quantum computing in the biomedical sciences as part of the Office of Data Science together with M. Madan Babu, PhD, Senior Vice President for Data Science, Chief Data Scientist, and Director of the Center of Excellence for Data-Driven Discovery and J. Paul Taylor, MD, PhD, St. Jude Executive Vice President and Scientific Director. 

“My lab will focus on developing next-generation methods for ligand and drug discovery, using quantum computing and machine learning to unify accuracy and speed when exploring the vast chemical space for ligand discovery,” Gorgulla explained. “It’s a natural fit for the Center of Excellence for Data Driven Discovery within the Structural Biology Department, as they continue to integrate large datasets and unravel protein structures and unique conformations they adopt. Our work can take the next important step of identifying ligands and develop them into tool compounds to initiate drug discovery efforts.”

Gorgulla began his work at St. Jude in 2023 as part of the Blue Sky for Seeing the Invisible in Protein Kinases, led by Charalampos Kalodimos, PhD, Structural Biology chair. There, he built a ligand discovery team that works on structure-based ligand discovery projects, looking at the conformation states (shapes) of protein kinases, the second most targeted proteins by pharmaceutical drugs. Through his work, he built research relationships and collaborated with colleagues, which paved the way for him to establish his own lab dedicated to transforming ligand discovery by harnessing quantum computing and machine learning.

“St. Jude has been an amazing fit for me scientifically,” Gorgulla concluded, “I’m happy to already be able to present early fruits of quantum computing’s application and hope to drive this wave of technology forward from the virtual and theoretical space to making a real-world difference for our patients.”