top of page
Search

Quantum-centric workflows for predicting RNA secondary structure

  • Writer: Chris Cade
    Chris Cade
  • May 20
  • 3 min read

In a recent collaboration with Moderna and IBM Quantum, we developed two quantum-centric optimization algorithms to predict mRNA secondary structure — a key challenge in computational biology with direct implications for mRNA-based drug design. The algorithms were implemented on IBM’s quantum processors, solving instances with up to 156 qubits. To explore scalability beyond the capability of current hardware, we used Fermioniq’s Ava to successfully solve a 354-qubit instance, one of the largest end-to-end simulations of a VQE algorithm ever performed.


Our results highlight the increasing potential of hybrid quantum-classical workflows for solving complex biological optimization tasks relevant to the pharmaceutical industry, and emphasise the critical role of advanced emulators like Ava in bridging today’s quantum capabilities with tomorrow’s quantum computers.



The challenge

Predicting how mRNA folds is critical for understanding gene expression and designing mRNA-based therapeutics, but remains a major challenge in computational biology due to the vast number of possible secondary structures.


To tackle this problem using quantum computers, the RNA folding task was formulated as a QUBO (quadratic unconstrained binary optimization) problem, a natural fit for quantum optimization and variational quantum algorithms. The QUBO problem was approached using two quantum-centric workflows:

  1. A Conditional Value at Risk (CVaR)-based variational quantum algorithm enhanced with gauge transformations and local search.

  2. Instantaneous Quantum Polynomial (IQP) circuit-based scheme where training is done classically and sampling is delegated to quantum hardware.


Fig. 1: a) Schematic for the quantum-centric architecture for sampling-based quantum optimization using CVaR. The training of the variational circuit is carried out by collecting samples from the hardware and processing them on classical compute nodes. b) Schematic of the approach based on the IQP circuit. Training of the variational circuit is conducted only on the classical compute nodes. The quantum processor is used for sampling from the optimized circuit.
Fig. 1: a) Schematic for the quantum-centric architecture for sampling-based quantum optimization using CVaR. The training of the variational circuit is carried out by collecting samples from the hardware and processing them on classical compute nodes. b) Schematic of the approach based on the IQP circuit. Training of the variational circuit is conducted only on the classical compute nodes. The quantum processor is used for sampling from the optimized circuit.

Both algorithms were run on IBM quantum processors, successfully solving problems instances with up to 156 qubits and 950 gates. To investigate scaling, tensor network simulations were used to solve problem instances with up to 354 qubits in ideal settings. These were performed using Fermioniq's Ava platform.

The project results provide a promising path towards solving longer mRNA sequences using a quantum-centric approach, and the experiments on real hardware not only highlight the growing capabilities of quantum hardware but also underscore the potential for significant advancements in understanding complex biological structures through quantum assisted methodologies



Pushing the limits of quantum emulation

Variational quantum algorithms require hundreds to thousands of repeated quantum circuit evaluations, demanding extremely fast emulation times and cost function evaluations. We optimized Ava to meet this challenge: for a 560-qubit problem, Ava simulated the circuit, performed over 32,000 measurements, and computed the CVaR cost function — all in under a second using a single NVIDIA GH200 GPU. Our performance optimizations allowed us to bring down the total time to simulate a full VQE algorithm from months to hours, allowing Ava to provide meaningful feedback during the quantum algorithm design process.


Ava's role in the collaboration highlights the value of high-performance quantum emulators for exploring the scalability of quantum algorithms far beyond the limits of today’s hardware, enabling meaningful investigation into their potential for industrial applications.




Acknowledgements

We would like to thank Moderna and IBM Quantum for the collaboration, and the team at Fermioniq for pushing Ava to even higher performance capabilities during this project.


For this research, Fermioniq was supported by the Dutch National Growth Fund (NGF), as part of the Quantum Delta NL Programme.

Read the full technical paper here.

bottom of page