Biotechnology · asia
From 3D Scaffolds to Experimental Feedback: RhoAgent Designs Functional RNA
A team at The Chinese University of Hong Kong has connected RNA structure generation, sequence design, and experimental results in an iterative refinement workflow, producing molecules capable of cleavage and fluorescence. The findings advance functional RNA engineering, but therapeutic applications remain some way off.
Beyond carrying genetic information, RNA can catalyze reactions and recognize molecules, serving as a material for cell engineering and biosensing. The challenge in making it perform a specified task is getting a string of nucleotides to fold into the right three-dimensional shape. RhoAgent, developed by a team at The Chinese University of Hong Kong, connects structure generation, sequence design, and experimental feedback. The team validated the functions of some designs in the laboratory. The study was published in *Nature Biotechnology*.
The system divides design work into four components—planning, execution, evaluation, and memory—and translates requirements expressed in natural language into molecular design constraints. Its core tool, RDiff, uses a two-stage diffusion model: it first generates an RNA backbone based on base interactions and spatial constraints on functional cores, then generates a compatible sequence. It can also preserve specified segments. Model training combines experimentally determined and predicted RNA structures, allowing design to start from three-dimensional geometry.
The team first tested ribozymes, RNA molecules that catalyze reactions. Some new designs shared only 40% to 70% sequence similarity with their corresponding natural sequences, yet retained comparable or even higher cleavage activity. This suggests that surrounding sequences may offer greater room for modification as long as key functional structures are maintained. However, these results come from experiments on selected molecules and do not imply that all generated sequences function properly.
A more challenging case was the hovlinc ribozyme, whose three-dimensional structure has not yet been experimentally determined. The team therefore used a predicted structure to guide design. Under the test conditions, the best candidate molecule had a cleavage fraction approximately six times that of an expert-designed truncated version; this benchmark was not the full-length natural ribozyme. After results from the first round of experiments were fed back into the system, the five best-performing variants in the second round had an average cleavage fraction 12% higher than in the first round, showing how experiments can guide the next round of searching.
Another validation focused on aptamers—RNA molecules that bind specific molecules. The team's designed Broccoli aptamer scaffold showed 230% higher fluorescence intensity than the natural version under the test conditions. They also placed the functional cores of two aptamers on the same RNA scaffold to build a fluorescence resonance energy transfer (FRET) system while preserving each aptamer's binding activity. This assembly capability could be explored in molecular sensing research because the signal depends on the spatial relationship between the functional cores.
These findings currently support the ability to design and optimize functional RNA; they have not yet demonstrated therapeutic efficacy. Cleavage fraction and fluorescence intensity are both measurements under specific experimental conditions and cannot be directly translated into performance in the human body. Moving toward medical applications will still require clarification of molecular stability in cellular environments, delivery, and safety. The university proposed integrating automated experimental platforms in the future. At this stage, RhoAgent's advance lies in subjecting designs to experimental testing and feeding the results back into the next round of refinement.