RNA design has become increasingly significant in modern medicine, particularly following advancements such as mRNA vaccines and gene therapies. On July 8, 2026, researchers from Keio University unveiled a novel RNA inverse folding framework that leverages artificial intelligence and an Ising machine to improve design efficiency.
Led by Project Lecturer Shuta Kikuchi and Professor Shu Tanaka, the team aimed to tackle the challenges of designing RNA molecules that reliably form desired secondary structures. Traditional computational methods often require extensive evaluations of potential candidates, creating bottlenecks that hinder progress.
Innovative Approach to RNA Design
The researchers developed a framework based on factorization machine with quadratic optimization annealing, referred to as FMQA. This approach is designed to identify high-quality RNA sequences with fewer evaluations compared to conventional methods. "We investigated a new application of FMQA in biomolecular design, where its potential remains relatively unexplored," said Dr. Kikuchi.
In their study, the researchers formulated RNA inverse folding as an optimization problem, focusing on identifying sequences likely to fold into a predefined target structure. FMQA was evaluated using four binary encoding methods: one-hot, domain-wall, binary, and unary.
Performance of Encoding Strategies
The study revealed that the choice of encoding strategy significantly impacts RNA design quality. The researchers assessed RNA design using the Normalized Ensemble Defect (NED), a metric that measures the accuracy of predicted structures against target structures.
- Encoding methods tested:
- One-hot
- Domain-wall
- Binary
- Unary
Results indicated that one-hot and domain-wall encodings consistently outperformed binary and unary representations. This led to sequences with lower NED values and greater success rates. Notably, domain-wall encoding introduced a search bias favoring specific integer states, enhancing the frequency of guanine (G) and cytosine (C) base pairs in stem regions, which ultimately improved thermodynamic stability.
Broader Implications for Computational Biology
The findings extend beyond RNA inverse folding, suggesting that annealing-based optimization frameworks like FMQA could be beneficial in various fields within computational biology and optimization science. The study emphasizes that data encoding is not just a preprocessing step but a critical design variable that can influence optimization outcomes.
Future applications of FMQA may include the design of functional biomolecules such as biosensors, genome-editing tools, and ribozymes. The flexibility of FMQA could also accommodate experimentally measured properties like molecular stability and binding affinity, enhancing the connection between computational design and laboratory validation.
As stated by Tanaka, "The insights gained from this study are not limited to RNA; they have a generality that allows them to be applied to discrete design problems where each evaluation is costly, including materials and molecular design." This perspective suggests a promising future for RNA design and other applications within biotechnology and medicine.
🤖 This article was rewritten by Feed and Figures' editorial AI from a report originally published by Phys.org. Facts and quotes are preserved from the original; the rewrite focuses on clarity and structure. For the unedited original, see the source link below.
