Keio University researchers have overcome a critical hurdle in RNA design by implementing a Factorization Machine with Quadratic Optimization Annealing (FMQA) framework, offering a more efficient alternative to conventional methods. The team, led by Project Lecturer Shuta Kikuchi and Professor Shu Tanaka, tackled the challenge of inverse folding, pinpointing RNA sequences that predictably fold into desired shapes, a process complicated by the exponential increase in possibilities with even short sequences. Their work, published in Scientific Reports on May 3, 2026, reveals that the method of converting RNA sequences into a binary format significantly impacts the success of AI-driven design. “We investigated a new application of FMQA in biomolecular design, where its potential remains relatively unexplored,” says Dr. Kikuchi, highlighting the study’s exploration of an underexplored dimension of biomolecular engineering.
A novel computational framework is reducing the time and cost associated with designing functional RNA molecules, offering a potential advance for therapeutics. Conventional methods struggle with the exponential growth of possibilities even in relatively short sequences, demanding extensive computational resources and experimental validation. Kikuchi found that how RNA sequences are converted into binary representations significantly influences optimization performance. Specifically, one-hot and domain-wall encodings consistently outperformed binary and unary methods, yielding sequences with improved thermodynamic stability. The researchers found that assigning guanine (G) and cytosine to favored states within domain-wall encoding encouraged G, C base pair formation in stem regions. Across benchmarks, FMQA identified high-quality RNA designs with fewer evaluations than competing methods like random search, genetic algorithms, and Bayesian optimization. Dr. Kikuchi suggests potential applications in biosensors, genome-editing tools, and aptamers, and the team anticipates that this evaluation-efficient approach will accelerate discovery across biotechnology and medicine, potentially utilizing quantum annealing machines in the future.
We investigated a new application of FMQA in biomolecular design, where its potential remains relatively unexplored. Since RNA, DNA, and protein sequences are inherently categorical in nature, it is unclear how converting them into binary representations affects optimization performance. In this study, we examined RNA inverse folding and the influence of different encoding and assignment choices within FMQA .
The pursuit of rationally designed RNA molecules, crucial for therapeutics, is increasingly reliant on artificial intelligence, but the method of translating biological sequences into a language AI can understand, the encoding strategy, has remained a surprisingly underexplored variable. Quality was assessed using the Normalized Ensemble Defect (NED), a measure of structural agreement. The findings revealed a clear hierarchy of performance. Dr. Kikuchi explains that one-hot and domain-wall encodings consistently outperformed their counterparts, generating sequences with lower NED values and higher success rates. This efficiency, coupled with the potential for implementation on quantum annealing machines, suggests a path toward accelerating biomolecular design and reducing the experimental burden on researchers. “They have a generality that allows them to be applied to discrete design problems where each evaluation is costly, including materials and molecular design,” adds Prof. Tanaka.
They have a generality that allows them to be applied to discrete design problems where each evaluation is costly, including materials and molecular design .
Keio University researchers are refining the metrics used to assess the quality of artificially designed RNA sequences, moving beyond simple accuracy to a more nuanced understanding of structural fidelity. This focus on structural agreement is critical, as even minor deviations can drastically alter a molecule’s function in therapeutic applications. Their work demonstrates that simply achieving a sequence is not enough; the quality of that sequence, as defined by NED, is paramount. Kikuchi further revealed a surprising influence of nucleotide assignment within the encoding scheme, highlighting that data encoding isn’t merely a preliminary step, but rather “a design variable that can fundamentally shape optimization outcomes.” The team’s findings suggest that a more holistic approach to RNA design, one that considers both the optimization algorithm and the encoding strategy, is essential for accelerating the development of functional biomolecules, from biosensors to gene-editing tools.
Potential applications include biosensors, genome-editing tools, aptamers, ribozymes, and riboswitches.
This success isn’t simply about the optimization algorithm itself; the researchers uncovered a previously “underexplored dimension of biomolecular engineering” through their investigation of encoding strategies. When guanine and cytosine were assigned to these favored states, the resulting RNA sequences exhibited greater thermodynamic stability, a critical factor for therapeutic applications. Looking ahead, the researchers envision a future where FMQA is integrated with quantum computing technologies. “Because FMQA formulates the learned surrogate model as a quadratic optimization problem, it can be implemented with quantum annealing machines,” says Dr. Kikuchi, pointing towards the potential of “advancing ‘Quantum for Biology’”. This convergence of quantum-inspired computing and biomolecular engineering could accelerate the design of functional biomolecules, biosensors, genome-editing tools, and aptamers, and ultimately reduce the experimental burden on biotechnology and medicine.
Because FMQA formulates the learned surrogate model as a quadratic optimization problem, it can be implemented with quantum annealing machines .
