Imagine trying to design a key for a lock that is constantly changing its shape. That is the exact challenge we face in modern drug discovery when dealing with intrinsically disordered proteins.
For decades, the classic “structure-function” paradigm in biology taught us that a protein’s amino acid sequence encodes a single, unique, and stable 3D structure, which in turn dictates what it does. However, nature is far more rebellious. A significant portion of the proteome—including roughly 79% of human cancer-associated proteins—defies this rule. These proteins contain intrinsically disordered regions (IDRs) that lack a stable folded structure under normal conditions. Instead of folding neatly, they exist as dynamic, shifting ensembles of conformations.
Because of their shapeshifting nature, IDRs have long been labeled “undruggable” by traditional structure-based drug design, which relies on stable 3D pockets to anchor small molecules. Even brilliant, revolutionary AI tools like AlphaFold2, trained on databases of strictly folded structures, struggle to capture the ensemble nature of these flexible regions.
Exhaustively searching the massive, flat energy landscape of an IDP requires traversing numerous degrees of freedom—an enormous computational task even for state-of-the-art supercomputers. This is where quantum computing offers a fresh advantage. By leveraging the principles of quantum mechanics, we can potentially explore these complex energy landscapes more effectively to find the lowest-energy conformations.
The paper is published in the journal PLOS One.
Quantum for biologists
My Ph.D. work set out to bridge the gap between abstract quantum algorithms and practical biological applications. We developed QuPepFold, a modular Python package specifically designed to democratize hybrid quantum-classical protein folding simulations.
Currently, there is a significant entry barrier for structural biologists who want to harness quantum resources but don’t have a background in quantum programming. QuPepFold acts as an automated abstraction layer. It hides the intimidating technical details of circuit construction and error mitigation, allowing researchers to simply input a peptide sequence and retrieve meaningful 3D structural data.
Here is how we do it under the hood: we map the discrete 3D conformational space of a peptide onto a tetrahedral lattice, where each amino acid’s position is specified by binary codes representing directional turns. To find the thermodynamic ground state—the protein’s native, most stable form—we use a Variational Quantum Eigensolver (VQE).
Filtering the noise
However, current quantum hardware is notoriously noisy. To combat this, we optimized our algorithm using a Conditional Value-at-Risk (CVaR) objective function. Instead of calculating the average energy of all the folds the quantum circuit predicts, the CVaR approach focuses strictly on the “tail” of the lowest-energy measurement results.
Moving forward
We designed QuPepFold to be hardware-agnostic; it currently runs seamlessly across Qiskit Aer simulators, Braket’s tensor-network simulator, and real physical devices like IonQ’s Aria-1 quantum computer. When we ran our simulations on the IonQ Aria-1, the algorithm successfully reproduced ground-state energies with over 90% fidelity. The momentum behind this project has been incredibly validating, especially following our recent acceptance into the IBM Qiskit ecosystem and securing AI Innovation Challenge infrastructure funding.
If you’re looking for hype, QuPepFold is the wrong story. This work does not “solve protein folding,” and it doesn’t magically make IDRs easy. It is a coarse-grained lattice-based approach targeted at very short peptides, and it still requires extensive sampling and has to contend with hardware noise.
Even conceptually, focusing on the ground state can miss biologically relevant ensemble diversity—an IDR’s function may depend on populated higher-energy states, not just the minimum. I see QuPepFold as a practical bridge: a reproducible way to experiment with quantum objectives (like CVaR) in a biophysically motivated pipeline today, while keeping the architecture modular enough to evolve as hardware and algorithms improve.