Science & Technology, Singapore (Commonwealth Union) – The Cancer Science Institute of Singapore (CSI Singapore) at the National University of Singapore (NUS) has achieved a breakthrough in RNA structure modeling by leveraging artificial intelligence (AI) and deep-learning techniques. Their innovative method, known as DRfold, surpasses traditional approaches, enhancing the accuracy of RNA 3D structure predictions by over 70 percent.
Led by Professor ZHANG Yang from CSI Singapore and the NUS School of Computing, the research team published their groundbreaking results in the scientific journal Nature Communications on September 16, 2023.
Researchers of the study pointed out that RNA, composed of single chains of nucleotides, is crucial for transcription and translation processes, facilitating the transfer of gene information from DNA to protein amino acid sequences. In recent years, the recognition of RNA’s vital roles in regulating biological processes has positioned it as a promising target for drug development.
Targeting RNAs with small molecules is anticipated to revolutionize drug design, offering an expansive landscape compared to the traditional focus on proteins. Consequently, the exploration of RNA biology and its applications in therapeutic development has become a pivotal and rapidly growing field, attracting substantial investment from both academic and industry sectors worldwide.
In contrast to well-folded protein structures, RNA structures and their folding patterns are generally considered less stable, primarily due to a relatively shallow energy landscape. Conventional physics- and statistics-based force fields, which are often prone to errors, struggle to precisely capture the intricate folding interactions of RNAs. The challenge is exacerbated by the limited availability of experimental RNA structures in the Protein Data Bank (PDB), constraining the accuracy of traditional knowledge-based force fields derived from PDB statistics.
To overcome these obstacles, DRfold has introduced two complementary deep-learning network pipelines. One is dedicated to end-to-end learning, while the other focuses on geometrical restraint learning. This innovative approach significantly enhances the accuracy of the AI-based force field. The synergy between these two networks further boosts the precision of the single neural network-based AI potentials.
Researchers of the study pointed out that the pivotal innovation lies in adopting a deep learning approach for predicting RNA tertiary structure. Unlike traditional methods relying on homologous modeling or physics-based folding simulations, which are limited by force field accuracy, DRfold employs self-attention transformer networks to predict 3D structures from RNA sequences. This marks a revolutionary shift in addressing the critical challenge of RNA structure prediction. DRfold’s novel strategy, incorporating two parallel and complementary networks based on end-to-end and geometry learnings, serves to improve the accuracy of the potential function and RNA model prediction. The result is a prediction method that is lightweight, highly flexible, scalable, and consequently, the preferred choice in addressing this intricate task.
Dr LI Yang, who is a Research Scientist at CSI Singapore as well as the first author of this study, says “Since the biological functions of RNAs depend on the specific tertiary structures, it becomes increasingly important and necessary to determine the 3D structures of RNAs in order to facilitate RNA-based function annotation and drug discovery.”
Dr Yang, also says “The golden standard in structural biology, such as using biophysical experiments — X-ray crystallography, Cryogenic Electron Microscopy (Cryo-EM), and Nuclear Magnetic Resonance (NMR) Spectroscopy — to determine RNA structures, are often cost- and labour-intensive, limiting their application to a tiny portion of known RNAs. Currently, there are more than 30 million known RNA sequences in the RNA central database, but only less than 500 (or 0.0017 per cent) have experimentally solved structures. This frustratingly leaves more than 99 per cent of RNA targets with no structural information. Hence, our study’s core aim is to develop new computational methods capable of predicting high-quality RNA structure models, filling this substantial information gap.”
