PINNACLE · Physics-Informed Neural Networks for Accurate Computational Learning of RNA Elements

Lines connect the coordinator with its partners.

Project objective

Ribonucleic acid (RNA) molecules are central to cellular function and disease, with their biological roles being intrinsically linked to their complex three-dimensional (3D) structures. However, a significant gap exists between the number of known RNA sequences and experimentally determined structures. This data scarcity severely hampers the application of state-of-the-art deep learning methods, which have revolutionized protein structure prediction but fail to generalize for RNA due to their reliance on vast datasets.The PINNACLE (Physics-Informed Neural Networks for Accurate Computational Learning of RNA Elements) project will address this fundamental challenge through a novel, physics-aware computational framework. The project has two primary objectives: 1) To construct and disseminate a high-quality, FAIR-compliant, and systematically curated dataset of RNA 3D structures, augmented with molecular simulations to capture molecular flexibility and rare interactions. 2) To develop and validate a novel Bayesian Physics-Informed Neural Network (B-PINN) that directly embeds the fundamental physical and biochemical laws governing RNA folding (e.g., electrostatics, base-stacking, torsional constraints) into the model's learning process.PINNACLE will reduce the dependency on large training sets, improve generalization to novel RNA families, and provide robust uncertainty quantification for its predictions - a critical feature for guiding experimental research. This action will deliver a transformative open-source tool for the scientific community, accelerating research in RNA biology and the development of RNA-based therapeutics. Furthermore, this interdisciplinary training will equip the researcher with a unique skill-set at the intersection of AI, biophysics, and bioinformatics, providing a robust foundation for a future independent research career.

Original text from CORDIS.

Participants

• KOBENHAVNS UNIVERSITET · KOBENHAVNCoordinatorDenmark
• STOCKHOLMS UNIVERSITET · StockholmSweden

Links

• View on CORDIS
• DOI: 10.3030/101284595