When Quantum Computers Start Designing Cancer Drugs
For decades, the KRAS protein has been one of cancer biology’s most frustrating puzzles. Mutated forms of KRAS drive some of the deadliest cancers—including pancreatic, lung, and colorectal tumors—yet the protein has long been considered “undruggable.” Its smooth surface and dynamic shape make it difficult for conventional drugs to latch onto, and despite years of effort, only a handful of KRAS-targeting medicines have reached the clinic.
Now, a team of researchers has taken a strikingly new approach: using quantum computing, alongside classical artificial intelligence, to design potential KRAS inhibitors from scratch. In a study published in Nature Biotechnology, they show that a hybrid quantum–classical algorithm can generate drug-like molecules that bind to KRAS—and that some of these molecules work in real biological experiments.
Why Drug Discovery Is So Hard
Finding a new drug is notoriously slow and expensive. The process often takes more than a decade and can cost billions of dollars, with around 90% of candidates failing along the way. One major challenge is the sheer size of “chemical space”: the number of possible drug-like molecules is estimated to be around 10⁶⁰. Even the most powerful classical computers can only explore a tiny fraction of that universe.
In recent years, AI-based “generative models” have helped by learning patterns from known drugs and proposing new molecules with desired properties. But even these models struggle when targets are especially complex—like KRAS.
This is where quantum computing enters the picture.
A Hybrid Quantum–Classical Strategy
The research team developed a hybrid system that combines a classical neural network with a quantum machine learning model known as a quantum circuit Born machine (QCBM). Quantum computers operate using qubits, which can exist in multiple states at once thanks to quantum effects such as superposition and entanglement. In theory, this allows them to represent complex probability distributions more efficiently than classical systems.
The workflow began with data. The researchers assembled a training set of more than 1.1 million molecules, starting with around 650 known KRAS inhibitors and expanding the pool using large-scale virtual screening and chemical space exploration. These molecules were then used to train the hybrid model, which generated entirely new candidate compounds designed to bind KRAS.
Using just 16 qubits on an IBM quantum processor, the quantum component provided a “prior” that guided the classical model toward higher-quality molecular designs.
From Computer to the Lab
After training, the team generated one million candidate molecules, which were filtered for drug-like properties and ranked by predicted binding strength. From this massive pool, 15 compounds were selected for synthesis and experimental testing.
Two molecules stood out.
One compound, called ISM061-018-2, showed particularly strong promise. Using surface plasmon resonance—a technique that measures molecular binding—the researchers found that it binds to the KRAS-G12D mutant with an affinity of 1.4 micromolar, a respectable result for such an early-stage molecule. In cell-based assays, ISM061-018-2 inhibited interactions between KRAS and its signaling partners across multiple cancer-relevant KRAS mutations, while showing no general toxicity to cells at concentrations up to 30 micromolar.
A second compound, ISM061-022, displayed a different but intriguing profile. It showed selective activity against certain KRAS mutants, including KRAS-G12R and KRAS-Q61H, suggesting that quantum-designed molecules could potentially be tuned for mutation-specific effects.
Did Quantum Computing Make a Difference?
To test whether the quantum component truly added value, the researchers compared their hybrid model with state-of-the-art classical approaches using a standard benchmarking platform. The results were striking: incorporating a quantum prior improved the success rate of generating viable molecules by about 21.5%, particularly for passing filters related to synthesizability and stability.
Even more intriguingly, the team found that performance improved roughly linearly with the number of qubits, hinting that larger quantum processors could further enhance molecular design in the future.
While the authors stop short of claiming a definitive “quantum advantage,” they argue that quantum effects likely helped the model explore regions of chemical space that classical algorithms struggle to reach.
What This Means for the Future of Medicine
This study marks one of the first demonstrations that quantum computing can contribute directly to experimentally validated drug discovery, rather than remaining a purely theoretical tool. The molecules identified are far from becoming medicines, but they provide new chemical starting points for tackling KRAS-driven cancers.
More broadly, the work suggests that hybrid quantum–classical approaches could shorten the early stages of drug discovery—from years to potentially months—by generating better candidates faster. As quantum hardware improves and qubit numbers grow, such methods may become increasingly powerful.
For now, the message is clear: quantum computers are no longer just abstract machines for physicists. They are beginning to design molecules that could one day become life-saving drugs.
The study is published in the journal Nature Biotechnology. It was led by Alex Zhavoronkov from Insilico Medicine AI limited, UAE.
