Abstract:RNA three-dimensional structure prediction has been relied on using a predicted or experimentally determined secondary structure as a restraint to reduce the conformational sampling space. However, the secondary-structure restraints are limited to paired bases, and the conformational space of the ribose-phosphate backbone is still too large to be sampled efficiently. Here, we employed the dilated convolutional neural network to predict backbone torsion and pseudotorsion angles using a single RNA sequence as input. The method called SPOT-RNA-1D was trained on a high-resolution training data set and tested on three independent, nonredundant, and high-resolution test sets. The proposed method yields substantially smaller mean absolute errors than the baseline predictors based on random predictions and based on helix conformations according to actual angle distributions. The mean absolute errors for three test sets range from 14°–44° for different angles, compared to 17°–62° by random prediction and 14°–58° by helix prediction. The method also accurately recovers the overall patterns of single or pairwise angle distributions. In general, torsion angles further away from the bases and associated with unpaired bases and paired bases involved in tertiary interactions are more difficult to predict. Compared to the best models in RNA-puzzles experiments, SPOT-RNA-1D yielded more accurate dihedral angles and, thus, are potentially useful as model quality indicators and restraints for RNA structure prediction as in protein structure prediction.The Supporting Information is available free of charge at <a class="ext-link" href="/doi/10.1021/acs.jcim.1c00153?goto=supporting-info">https://pubs.acs.org/doi/10.1021/acs.jcim.1c00153</a>.Table S1, number of RNAs, base pairs, and length distribution of data sets; Table S2, statistical significance of SPOT-RNA-1D improvement over random-baseline predictor; Table S3, performance comparison SPOT-RNA-1D and random-baseline predictor based on helix angle distribution; Table S4, performance comparison of SPOT-RNA-1D with RNA-Puzzles predictors; Table S5, performance comparison of method with single-sequence input and with single-sequence plus predicted secondary structure; Figure S1, distribution plots of native torsion angles of training data; Figure S2, distribution plots of native and predicted torsion angles; Figure S3, distribution plots of native and predicted pseudotorsion angles; Figure S4, MAE vs RMSD scatterplot of nine RNAs from RNAPOT; Figure S5, MAE vs GDT-score scatterplot of nine RNAs from RNAPOT; Figure S6, MAE between native angles and angle from RNAPOT models for 1kxk-A as function of RMSD; Figure S7, MAE between native angles and angle from RNAPOT models for 1mzp-B as function of RMSD; Figure S8, MAE between native angles and angle from RNAPOT models for 1s03-A as function of RMSD; Figure S9, MAE between native angles and angle from RNAPOT models for 1u63-Chain B as function of RMSD; Figure S10, MAE between native angles and angle from RNAPOT models for 1un6-E as function of RMSD; Figure S11, MAE between native angles and angle from RNAPOT models for 2dr2-B as function of RMSD; Figure S12, MAE between native angles and angle from RNAPOT models for 2oiu-P as function of RMSD; Figure S13, MAE between native angles and angle from RNAPOT models for 2qwy-A as function of RMSD; Figure S14, MAE between native angles and angle from RNAPOT models for 2zni-C as function of RMSD; Figure S15, 3D structural alignment of native structure and decoy-17 of 2dr2-B; and Figure S16, distribution plots of native and predicted torsion angles by method with single sequence and predicted secondary structure as input (<a class="ext-link" href="/doi/suppl/10.1021/acs.jcim.1c00153/suppl_file/ci1c00153_si_001.pdf">PDF</a>)This article has not yet been cited by other publications.

Deep Learning Methods for RNA Torsion Angle Prediction

RNA Backbone Torsion and Pseudotorsion Angle Prediction Using Dilated Convolutional Neural Networks

A Hopfield Neural Network Based Algorithm for RNA Secondary Structure Prediction

Deep learning methods for protein torsion angle prediction

RNA-TorsionBERT: leveraging language models for RNA 3D torsion angles prediction

A Comparative Review of Deep Learning Methods for RNA Tertiary Structure Prediction

Systematic benchmarking of deep-learning methods for tertiary RNA structure prediction

DeepRNA-Twist: Language Model guided RNA Torsion Angle Prediction with Attention-Inception Network

TorRNA - Improved Prediction of Backbone Torsion Angles of RNA by Leveraging Large Language Models

Improvements of the Hierarchical Approach for Predicting Rna Tertiary Structure

Computational Prediction of RNA Tertiary Structures Using Machine Learning Methods*

Accurate RNA 3D Structure Prediction Using a Language Model-Based Deep Learning Approach

Prediction of the RNA Tertiary Structure Based on a Random Sampling Strategy and Parallel Mechanism

RNA Secondary Structure Prediction Using Transformer-Based Deep Learning Models

RNAformer: A Simple yet Effective Model for Homology-Aware RNA Secondary Structure Prediction

Simultaneous prediction of RNA secondary structure and helix coaxial stacking

Machine learning for RNA 2D structure prediction benchmarked on experimental data

Methods and Applications of RNA Contact Prediction*

RNA secondary structure prediction using deep learning with thermodynamic integration

Accurate prediction of protein torsion angles using evolutionary signatures and recurrent neural network

Ab initio Prediction of RNA Nucleotide Interactions with Backbone k-Tree Model