RNA3DCNN: Local and global quality assessments of RNA 3D structures using 3D deep convolutional neural networks

Abstract

Quality assessment is essential for the computational prediction and design of RNA tertiary structures. To date, several knowledge-based statistical potentials have been proposed and proved to be effective in identifying native and near-native RNA structures. All these potentials are based on the inverse Boltzmann formula, while differing in the choice of the geometrical descriptor, reference state, and training dataset. Via an approach that diverges completely from the conventional statistical potentials, our work explored the power of a 3D convolutional neural network (CNN)-based approach as a quality evaluator for RNA 3D structures, which used a 3D grid representation of the structure as input without extracting features manually. The RNA structures were evaluated by examining each nucleotide, so our method can also provide local quality assessment. Two sets of training samples were built. The first one included 1 million samples generated by high-temperature molecular dynamics (MD) simulations and the second one included 1 million samples generated by Monte Carlo (MC) structure prediction. Both MD and MC procedures were performed for a non-redundant set of 414 RNAs. For two training datasets (one including only MD training samples and the other including both MD and MC training samples), we trained two neural networks, named RNA3DCNN_MD and RNA3DCNN_MDMC, respectively. The former is suitable for assessing near-native structures, while the latter is suitable for assessing structures covering large structural space. We tested the performance of our method and made comparisons with four other traditional scoring functions. On two of three test datasets, our method performed similarly to the state-of-the-art traditional scoring function, and on the third test dataset, our method was far superior to other scoring functions. Our method can be downloaded from https://github.com/lijunRNA/RNA3DCNN.

Salt-Dependent RNA Pseudoknot Stability: Effect of Spatial Confinement

“…Although experimental methods, such as X-ray crystallography, nuclear magnetic resonance spectroscopy, and cryo-electron microscopy, can be used to determine the structures of RNAs including pseudoknots, the structures in Protein Data Bank (PDB; https://www.rcsb.org) are still limited due to the high cost of the experimental measurements (Hajdin et al, 2010;Rose et al, 2011;Shi et al, 2014b;Schlick and Pyle, 2017). To complement the experiments, some computational models/methods (e.g., FARNA, MC-Fold/MC-Sym, Vfold, iFoldRNA, 3dRNA, RNAComposer, SimRNA, oxRNA, HiRE-RNA, and pk3D) have been developed for predicting RNA 3D structures (Cao and Chen, 2005;Ding et al, 2008;Parisien and Major, 2008;Zhang et al, 2009;Das et al, 2010;Popenda et al, 2012;Zhao et al, 2012;He et al, 2013He et al, , 2015He et al, , 2019Kim et al, 2014;Liwo et al, 2014Liwo et al, , 2020Sulc et al, 2014;Cragnolini et al, 2015;Wang et al, 2015a,b;Boniecki et al, 2016;Dawson et al, 2016;Li et al, 2016Li et al, , 2018Tan et al, 2019). Most of these models/methods are primarily designed to predict folded structures and cannot predict the stability of RNAs, especially in ion solutions (Shi et al, 2014b;Dawson et al, 2016;Schlick and Pyle, 2017), whereas the structural stability of RNAs can be very sensitive to ion conditions due to their polyanionic nature (Das et al, 2005;Draper et al, 2005;Chen, 2007, 2011;Qiu et al, 2010;Lipfert et al, 2014;Wang et al, 2018<...>…”

What is the best reference state for building statistical potentials in RNA 3D structure evaluation?

“…Moreover, a multibody statistical potential (Singh et al 1996;Feng et al 2007;Masso 2018) can possibly capture more structural features than conventional pairwise ones, while generally involving a higher computational cost. Finally, machine-learning methods can be applied in building the statistical potential to dig critical information not easily detected for RNA structures (Li et al 2018).…”

Machine Learning Approaches on High Throughput NGS Data to Unveil Mechanisms of Function in Biology and Disease

“…ML techniques can be employed to learn the 3D properties such as dihedral angles and total energies per cluster of molecules. One such method is RNA3DCNN (61), where the RNA molecules can be treated as a 3D image or voxels as input to 3D CNNs. The RNA molecule is described using a 3D grid representation of the RNA molecules on a cartesian coordinate system directly as input to the convolutional neural network.…”