UFold

UFold predicts RNA secondary structure from nucleotide sequences using deep learning to improve accuracy and speed of structure inference.

Key Features:

• Deep Learning Approach: Employs a deep learning architecture trained on annotated RNA structures and does not rely on thermodynamic free-energy assumptions.
• Performance Improvements: Reports approximately 10–30% improvement over traditional thermodynamic models, a ~14% improvement over other learning-based approaches, and a base-pair prediction F1 score of 0.91 on benchmark datasets.
• Speed: Processes sequences up to 1600 base pairs in approximately 160 milliseconds per sequence.

Scientific Applications:

• Gene regulation: Supports investigation of RNA-mediated gene regulation by providing predicted secondary structures.
• RNA–protein interactions: Aids study of RNA–protein interactions through predicted structural contexts of binding sites.
• Non-coding RNAs: Facilitates characterization of non-coding RNA structure–function relationships.
• Disease and therapeutics: Supports research into RNA-related diseases and the design of RNA-based therapeutics.
• Cellular processes: Enhances understanding of RNA roles in cellular processes by supplying structural hypotheses.

Methodology:

A deep learning model trained on annotated RNA secondary-structure datasets learns sequence-to-structure patterns without thermodynamic priors.