Near-native Conformations Research Articles

Characterizing the Conformational Dynamics of Human SUMO2: Insights into its Interaction with Metal Ions and SIMs.

SUMO (Small Ubiquitin-like Modifiers) proteins are involved in a crucial post-translational modification commonly termed as SUMOylation. In this work, we have investigated the native-state conformational flexibility of human SUMO2 and its interaction with Cu2+ and Zn2+ ions using 15N-1H based 2D NMR spectroscopy. After SUMO1, SUMO2 is the most studied SUMO isoform in humans which shares 45 % and ~80 % similarity with SUMO1 in terms of sequence and structure, respectively. In this manuscript, we demonstrate that compared to SUMO1, several amino acids around the α1-helix region of SUMO2 access energetically similar near-native conformations. These conformations could play a crucial role in SUMO2's non-covalent interactions with SUMO interaction motifs (SIMs) on other proteins. The C-terminal of SUMO2 was found to bind strongly with Cu2+ ions resulting in a trimeric structure as observed by gel electrophoresis. This interaction seems to interfere in its non-covalent interaction with a V/I-x-V/I-V/I based SIM in Daxx protein.

GSScore: a novel Graphormer-based shell-like scoring method for protein-ligand docking.

Protein-ligand interactions (PLIs) are essential for cellular activities and drug discovery. But due to the complexity and high cost of experimental methods, there is a great demand for computational approaches to recognize PLI patterns, such as protein-ligand docking. In recent years, more and more models based on machine learning have been developed to directly predict the root mean square deviation (RMSD) of a ligand docking pose with reference to its native binding pose. However, new scoring methods are pressingly needed in methodology for more accurate RMSD prediction. We present a new deep learning-based scoring method for RMSD prediction of protein-ligand docking poses based on a Graphormer method and Shell-like graph architecture, named GSScore. To recognize near-native conformations from a set of poses, GSScore takes atoms as nodes and then establishes the docking interface of protein-ligand into multiple bipartite graphs within different shell ranges. Benefiting from the Graphormer and Shell-like graph architecture, GSScore can effectively capture the subtle differences between energetically favorable near-native conformations and unfavorable non-native poses without extra information. GSScore was extensively evaluated on diverse test sets including a subset of PDBBind version 2019, CASF2016 as well as DUD-E, and obtained significant improvements over existing methods in terms of RMSE, $R$ (Pearson correlation coefficient), Spearman correlation coefficient and Docking power.

ViTScore: a novel three-dimensional Vision Transformer method for accurate prediction of protein-ligand docking poses.

Protein-ligand interactions (PLIs) are essential for cellular activities and drug discovery, and due to the complexity and high cost of experimental methods, there is a great demand for computational approaches, such as protein-ligand docking, to decipher PLI patterns. One of the most challenging aspects of protein-ligand docking is to identify near-native conformations from a set of poses, but traditional scoring functions still have limited accuracy. Therefore, new scoring methods are urgently needed for methodological and/or practical implications. We present a novel deep learning-based scoring function for ranking protein-ligand docking poses based on Vision Transformer (ViT), named ViTScore. To recognize near-native poses from a set of poses, ViTScore voxelizes the protein-ligand interactional pocket into a 3D grid labeled by the occupancy contribution of atoms in different physicochemical classes. This allows ViTScore to capture the subtle differences between spatially and energetically favorable near-native poses and unfavorable non-native poses without needing extra information. After that, ViTScore will output the prediction of the root mean square deviation (rmsd) of a docking pose with reference to the native binding pose. ViTScore is extensively evaluated on diverse test sets including PDBbind2019 and CASF2016, and obtains significant improvements over existing methods in terms of RMSE, R and docking power. Moreover, the results demonstrate that ViTScore is a promising scoring function for protein-ligand docking, and it can be used to accurately identify near-native poses from a set of poses. Furthermore, the results suggest that ViTScore is a powerful tool for protein-ligand docking, and it can be used to accurately identify near-native poses from a set of poses. Additionally, ViTScore can be used to identify potential drug targets and to design new drugs with improved efficacy and safety.

Number of Hydrogen Bonds per Unit Solvent Accessible Surface Area: A Descriptor of Functional States of Proteins.

Proteins function close to native and near-native conformations. These states are evolutionarily selected to ensure the effect of mutations is minimized. The structural organization of a protein is hierarchical and modular, which reduces the dimensionality of the configurational space of the native states. Thus, finding appropriate descriptors that define the native state among all possible states of a protein is a problem of immense interest. The present study explores the correlation between solvent accessible surface areas (SASAs) and different intraprotein as well as protein-water hydrogen bonds of 55 single-chain globular proteins from four different structural classes (all α, all β, α+β, and α/β), 16 multichain proteins, and 4 macromolecular complexes. A systematic analysis of the solvent accessible surface area and intraprotein and protein-water hydrogen bonds suggests a linear relationship between SASAs and hydrogen bonds. The number of protein-water hydrogen bonds per unit SASA ranges from 3 to 4 for all the different structural protein classes. In contrast, the number of intramolecular hydrogen bonds per unit SASA, including the mainchain-mainchain, mainchain-sidechain, and sidechain-sidechain, varies between 0.75 to 2. The solvation free energy of a protein linearly decreases with SASA. Our study also shows that the solvation free energy/SASA varies from -75 to -105 kJ mol-1 nm-2 across all the native states studied here. The number conservancy of intraprotein hydrogen bonds per unit SASA possibly imparts structural stability to the native structure. On the other hand, 3-4 protein-water hydrogen bonds per unit SASA are possibly required to maintain a balance between the solubility and functionality of the native states. This study provides a basis for synthetic biologists to design new folds with improved functionality.

Deep Local Analysis evaluates protein docking conformations with locally oriented cubes.

MotivationWith the recent advances in protein 3D structure prediction, protein interactions are becoming more central than ever before. Here, we address the problem of determining how proteins interact with one another. More specifically, we investigate the possibility of discriminating near-native protein complex conformations from incorrect ones by exploiting local environments around interfacial residues.ResultsDeep Local Analysis (DLA)-Ranker is a deep learning framework applying 3D convolutions to a set of locally oriented cubes representing the protein interface. It explicitly considers the local geometry of the interfacial residues along with their neighboring atoms and the regions of the interface with different solvent accessibility. We assessed its performance on three docking benchmarks made of half a million acceptable and incorrect conformations. We show that DLA-Ranker successfully identifies near-native conformations from ensembles generated by molecular docking. It surpasses or competes with other deep learning-based scoring functions. We also showcase its usefulness to discover alternative interfaces.Availability and implementation http://gitlab.lcqb.upmc.fr/dla-ranker/DLA-Ranker.git Supplementary information Supplementary data are available at Bioinformatics online.

Efficient 3D conformer generation of cyclic peptides formed by a disulfide bond

Cyclic peptides formed by disulfide bonds have been one large group of common drug candidates in drug development. Structural information of a peptide is essential to understand its interaction with its target. However, due to the high flexibility of peptides, it is difficult to sample the near-native conformations of a peptide. Here, we have developed an extended version of our MODPEP approach, named MODPEP2.0, to fast generate the conformations of cyclic peptides formed by a disulfide bond. MODPEP2.0 builds the three-dimensional (3D) structures of a cyclic peptide from scratch by assembling amino acids one by one onto the cyclic fragment based on the constructed rotamer and cyclic backbone libraries. Being tested on a data set of 193 diverse cyclic peptides, MODPEP2.0 obtained a considerable advantage in both accuracy and computational efficiency, compared with other sampling algorithms including PEP-FOLD, ETKDG, and modified ETKDG (mETKDG). MODPEP2.0 achieved a high sampling accuracy with an average Calpha RMSD of 2.20 Å and 1.66 Å when 10 and 100 conformations were considered, respectively, compared with 3.41 Å and 2.62 Å for PEP-FOLD, 3.44 Å and 3.16 Å for ETKDG, 3.09 Å and 2.72 Å for mETKDG. MODPEP2.0 also reproduced experimental peptide structures for 81.35% of the test cases when an ensemble of 100 conformations were considered, compared with 54.95%, 37.50% and 50.00% for PEP-FOLD, ETKDG, and mETKDG. MODPEP2.0 is computationally efficient and can generate 100 peptide conformations in one second. MODPEP2.0 will be useful in sampling cyclic peptide structures and modeling related protein-peptide interactions, facilitating the development of cyclic peptide drugs.

TRScore: a 3D RepVGG-based scoring method for ranking protein docking models.

Protein-protein interactions (PPI) play important roles in cellular activities. Due to the technical difficulty and high cost of experimental methods, there are considerable interests towards the development of computational approaches, such as protein docking, to decipher PPI patterns. One of the important and difficult aspects in protein docking is recognizing near-native conformations from a set of decoys, but unfortunately, traditional scoring functions still suffer from limited accuracy. Therefore, new scoring methods are pressingly needed in methodological and/or practical implications. We present a new deep learning-based scoring method for ranking protein-protein docking models based on a 3D RepVGG network, named TRScore. To recognize near-native conformations from a set of decoys, TRScore voxelizes the protein-protein interface into a 3D grid labeled by the number of atoms in different physicochemical classes. Benefiting from the deep convolutional RepVGG architecture, TRScore can effectively capture the subtle differences between energetically favorable near-native models and unfavorable non-native decoys without needing extra information. TRScore was extensively evaluated on diverse test sets including protein-protein docking benchmark 5.0 update set, DockGround decoy set, as well as realistic CAPRI decoy set and overall obtained a significant improvement over existing methods in cross-validation and independent evaluations. Codes available at: https://github.com/BioinformaticsCSU/TRScore.

LEADS-FRAG: A Benchmark Data Set for Assessment of Fragment Docking Performance.

Fragment-based drug design is a popular approach in drug discovery, which makes use of computational methods such as molecular docking. To assess fragment placement performance of molecular docking programs, we constructed LEADS-FRAG, a benchmark data set containing 93 high-quality protein-fragment complexes that were selected from the Protein Data Bank using a rational and unbiased process. The data set contains fully prepared protein and fragment structures and is publicly available. Moreover, we used LEADS-FRAG for evaluating the small-molecule docking programs AutoDock, AutoDock Vina, FlexX, and GOLD for their fragment docking performance. GOLD in combination with the scoring function ChemPLP and AutoDock Vina performed best and generated near-native conformations (root mean square deviation <1.5 Å) for more than 50% of the data set considering the top-ranked docking pose. Taking into account all docking poses, the tested programs generated near-native conformations for up to 86% of the fragments in LEADS-FRAG. By rescoring all docking poses with the GOLD scoring functions and the Protein-Ligand Informatics force field, the number of near-native conformations increased up to 40% with respect to the top-rescored poses. Our results show that conventional small-molecule docking programs achieve a satisfactory fragment docking performance when utilizing rescoring.

Partial Least-Squares Discriminant Analysis and Ensemble-Based Flexible Docking of PD-1/PD-L1 Inhibitors: A Pilot Study.

Although mAbs targeting the programmed cell death protein 1 (PD-1)/programmed cell death ligand 1 (PD-L1) pathway have achieved remarkable therapeutic potential against multiple types of cancer, it is still of great interest for researchers to develop small-molecule PD-1/PD-L1 inhibitors without the mAb-related disadvantages of no oral bioavailability and poor solid tumor penetration. However, targeting the PD-1/PD-L1 pathway with small molecules is normally considered challenging because of the flat and large interaction surface of the PD-1/PD-L1 complex. In this paper, a total of 2558 PD-1/PD-L1 inhibitors were compiled from recent patents and literatures and then used for exploring the chemical space and structural features of PD-1/PD-L1 inhibitors by partial least-squares discriminant analysis. The results showed that intramolecular H bond, amphotericity indices, radius of gyration, nonbond electrostatic energy, fractional van der Waals surface area of H-bond donors, octanol–water partition coefficient, and molecular weight are the seven key features discriminating the PD-1/PD-L1 inhibitors from noninhibitors, with the prediction accuracy larger than 0.90. Based on the seven crystal structures of the PD-L1 dimer complexed with the patent BristolMyers Squibb (BMS) inhibitors, the feasibility of molecular docking for this unconventional binding pocket was further investigated. The results showed that the ensemble-based flexible docking protocol can reproduce the near-native binding conformations of the BMS inhibitors with a strong correlation between the IC50 values and ligand–receptor interaction energies (R = 0.81). In general, this paper delineates, for the first time, the characteristic features of the PD-1/PD-L1 inhibitors as well as a high-quality flexible docking strategy for the unconventional binding pocket of the PD-L1 dimer.

Big Data from Sparse Data: Diverse Scientific Benchmarks Reveal Optimization Imperatives for Implicit Membrane Energy Functions

In computational protein structure prediction and design, energy functions discriminate non-native from near-native biomolecule conformations. These functions approximate the balance of enthalpic and entropic contributions to protein stability through mathematical models derived from thermodynamic and structural data. Over the past decade, an influx of high-resolution data paired with machine learning techniques has significantly improved the accuracy of soluble protein energy functions. However, membrane protein energy functions remain low-resolution because the experimental data are sparse, leading to overfitting. To overcome this challenge, we assembled a suite of 12 computational benchmark tests against experimental targets. The tests probe various membrane protein energy function capabilities ranging from reproducing protein stabilities and orientations to accurately predicting the three-dimensional structures of monomeric proteins and protein complexes. To evaluate current performance, we ran the benchmark on Franklin19, the current state-of-the-art implicit membrane protein energy function in Rosetta. The benchmark revealed areas for improvement including treatment of electrostatic interactions, representation of the interfacial head-group region, and accounting for entropic effects on protein orientation. Ultimately, these benchmarks will enable big-data-style optimization of membrane energy functions, leading to improved membrane protein design capabilities.

Coupling enhanced sampling of the apo-receptor with template-based ligand conformers selection: performance in pose prediction in the D3R Grand Challenge 4.

We report the performance of our newly introduced Ensemble Docking with Enhanced sampling of pocket Shape (EDES) protocol coupled to a template-based algorithm to generate near-native ligand conformations in the 2019 iteration of the Grand Challenge (GC4) organized by the D3R consortium. Using either AutoDock4.2 or HADDOCK2.2 docking programs (each software in two variants of the protocol) our method generated native-like poses among the top 5 submitted for evaluation for most of the 20 targets with similar performances. The protein selected for GC4 was the human beta-site amyloid precursor protein cleaving enzyme 1 (BACE-1), a transmembrane aspartic-acid protease. We identified at least one pose whose heavy-atoms RMSD was less than 2.5Å from the native conformation for 16 (80%) and 17 (85%) of the 20 targets using AutoDock and HADDOCK, respectively. Dissecting the possible sources of errors revealed that: (i) our EDES protocol (with minor modifications) was able to sample sub-ångstrom conformations for all 20 protein targets, reproducing the correct conformation of the binding site within ~ 1Å RMSD; (ii) as already shown by some of us in GC3, even in the presence of near-native protein structures, a proper selection of ligand conformers is crucial for the success of ensemble-docking calculations. Importantly, our approach performed best among the protocols exploiting only structural information of the apo protein to generate conformations of the receptor for ensemble-docking calculations.

Accelerated Molecular Dynamics Applied to the Peptaibol Folding Problem.

The use of enhanced sampling molecular dynamics simulations to facilitate the folding of proteins is a relatively new approach which has quickly gained momentum in recent years. Accelerated molecular dynamics (aMD) can elucidate the dynamic path from the unfolded state to the near-native state, “flattened” by introducing a non-negative boost to the potential. Alamethicin F30/3 (Alm F30/3), chosen in this study, belongs to the class of peptaibols that are 7–20 residue long, non-ribosomally synthesized, amphipathic molecules that show interesting membrane perturbing activity. The recent studies undertaken on the Alm molecules and their transmembrane channels have been reviewed. Three consecutive simulations of ~900 ns each were carried out where N-terminal folding could be observed within the first 100 ns, while C-terminal folding could only be achieved almost after 800 ns. It took ~1 μs to attain the near-native conformation with stronger potential boost which may take several μs worth of classical MD to produce the same results. The Alm F30/3 hexamer channel was also simulated in an E. coli mimicking membrane under an external electric field that correlates with previous experiments. It can be concluded that aMD simulation techniques are suited to elucidate peptaibol structures and to understand their folding dynamics.

Improved fragment sampling for ab initio protein structure prediction using deep neural networks

A typical approach to predicting unknown native structures of proteins is to assemble the amino acid residues (fragments) extracted from known structures. The quality of these extracted fragments, which are used to build protein-specific fragment libraries, can determine the success or failure of sampling near-native conformations. Here we show how a high-quality fragment library can be built using deep contextual learning techniques. Our algorithm, called DeepFragLib, employs bidirectional long short-term-memory recurrent neural networks with knowledge distillation for initial fragment classification, followed by an aggregated residual transformation network with cyclically dilated convolution for detecting near-native fragments. DeepFragLib improves the position-averaged proportion of near-native fragments by 12.2% over existing methods and, consequently, produces better near-native structures for 72.0% of the free-modelling domain targets tested when integrated with Rosetta. DeepFragLib is fully parallelized and available for use in conjunction with structure prediction programs. An approach to protein structure prediction is to assemble candidate structures from template fragments, which are extracted from known protein structures. Wang et al. demonstrate that combining deep neural network architectures with a relatively small but high-resolution fragment dataset can improve the quality of the sample fragment libraries used for protein structure prediction.

IScore: a novel graph kernel-based function for scoring protein-protein docking models.

MotivationProtein complexes play critical roles in many aspects of biological functions. Three-dimensional (3D) structures of protein complexes are critical for gaining insights into structural bases of interactions and their roles in the biomolecular pathways that orchestrate key cellular processes. Because of the expense and effort associated with experimental determinations of 3D protein complex structures, computational docking has evolved as a valuable tool to predict 3D structures of biomolecular complexes. Despite recent progress, reliably distinguishing near-native docking conformations from a large number of candidate conformations, the so-called scoring problem, remains a major challenge.ResultsHere we present iScore, a novel approach to scoring docked conformations that combines HADDOCK energy terms with a score obtained using a graph representation of the protein–protein interfaces and a measure of evolutionary conservation. It achieves a scoring performance competitive with, or superior to, that of state-of-the-art scoring functions on two independent datasets: (i) Docking software-specific models and (ii) the CAPRI score set generated by a wide variety of docking approaches (i.e. docking software-non-specific). iScore ranks among the top scoring approaches on the CAPRI score set (13 targets) when compared with the 37 scoring groups in CAPRI. The results demonstrate the utility of combining evolutionary, topological and energetic information for scoring docked conformations. This work represents the first successful demonstration of graph kernels to protein interfaces for effective discrimination of near-native and non-native conformations of protein complexes.Availability and implementationThe iScore code is freely available from Github: https://github.com/DeepRank/iScore (DOI: 10.5281/zenodo.2630567). And the docking models used are available from SBGrid: https://data.sbgrid.org/dataset/684).Supplementary information Supplementary data are available at Bioinformatics online.

Methods for the Refinement of Protein Structure 3D Models.

The refinement of predicted 3D protein models is crucial in bringing them closer towards experimental accuracy for further computational studies. Refinement approaches can be divided into two main stages: The sampling and scoring stages. Sampling strategies, such as the popular Molecular Dynamics (MD)-based protocols, aim to generate improved 3D models. However, generating 3D models that are closer to the native structure than the initial model remains challenging, as structural deviations from the native basin can be encountered due to force-field inaccuracies. Therefore, different restraint strategies have been applied in order to avoid deviations away from the native structure. For example, the accurate prediction of local errors and/or contacts in the initial models can be used to guide restraints. MD-based protocols, using physics-based force fields and smart restraints, have made significant progress towards a more consistent refinement of 3D models. The scoring stage, including energy functions and Model Quality Assessment Programs (MQAPs) are also used to discriminate near-native conformations from non-native conformations. Nevertheless, there are often very small differences among generated 3D models in refinement pipelines, which makes model discrimination and selection problematic. For this reason, the identification of the most native-like conformations remains a major challenge.

The SIRAH 2.0 Force Field: Altius, Fortius, Citius.

A new version of the coarse-grained (CG) SIRAH force field for proteins has been developed. Modifications to bonded and non-bonded interactions on the existing molecular topologies significantly ameliorate the structural description and flexibility of a non-redundant set of proteins. The SIRAH 2.0 force field has also been ported to the popular simulation package AMBER, which along with the former implementation in GROMACS expands significantly the potential range of users and performance of this CG force field on CPU/GPU codes. As a non-trivial example of its application, we undertook the structural and dynamical analysis of the most abundant and conserved calcium-binding protein, calmodulin (CaM). CaM is composed of two calcium-binding motifs called EF-hands, which in the presence of calcium specifically recognize a cognate peptide by embracing it. CG simulations of CaM bound to four calcium ions in the presence or absence of a binding peptide (holo and apo forms, respectively) resulted in good and stable ion coordination. The simulation of the holo form starting from an experimental structure sampled near-native conformations, retrieving quasi-atomistic precision. Removing the binding peptide enabled the EF-hands to perform large reciprocal movements, comparable to those observed in NMR structures. On the other hand, the isolated peptide starting from the helical conformation experienced spontaneous unfolding, in agreement with previous experimental data. However, repositioning the peptide in the neighborhood of one EF-hand not only prevented the peptide from unfolding but also drove CaM to a fully bound conformation, with both EF-hands embracing the cognate peptide, resembling the experimental holo structure. Therefore, SIRAH 2.0 shows the capacity to handle a number of structurally and dynamically challenging situations, including metal ion coordination, unbiased conformational sampling, and specific protein-peptide recognition.

A Physical Modeling Approach to Determine Protein Structures from Paramagnetic NMR Measurements

Paramagnetic NMR approaches to protein structure determination can provide critical long-range information that may be difficult to obtain using standard NOE measurements, particularly for larger proteins. The paramagnetic relaxation enhancement (PRE) effect can be used to infer distances between residues and site-specifically engineered paramagnetic spin labels. These measurements are both sparse, due to the cost of making spin-labeled mutants, and ambiguous, due to the inherent flexibility of the spin-label. Here, we demonstrate the application of MELD (Modeling Employing Limited Data), a Bayesian approach combing sparse and ambiguous data with physical modeling, to the structure determination of a Calmodulin-peptide complex. Ten single-cysteine mutants were labeled with the common MTSL spin label, and PRE measurements were taken. Our approach can identify dominant conformations within 4 Å RMSD of the reference crystal structure using an implicit solvent model, while simulations without PRE data yield no near-native conformations. Inclusion of an explicit solvent significantly improves the model in comparison to the reference.

Hierarchical Flexible Peptide Docking by Conformer Generation and Ensemble Docking of Peptides.

Given the importance of peptide-mediated protein interactions in cellular processes, protein-peptide docking has received increasing attention. Here, we have developed a Hierarchical flexible Peptide Docking approach through fast generation and ensemble docking of peptide conformations, which is referred to as HPepDock. Tested on the LEADS-PEP benchmark data set of 53 diverse complexes with peptides of 3-12 residues, HPepDock performed significantly better than the 11 docking protocols of five small-molecule docking programs (DOCK, AutoDock, AutoDock Vina, Surflex, and GOLD) in predicting near-native binding conformations. HPepDock was also evaluated on the 19 bound/unbound and 10 unbound/unbound protein-peptide complexes of the Glide SP-PEP benchmark and showed an overall better performance than Glide SP-PEP+MM-GBSA and FlexPepDock in both bound and unbound docking. HPepDock is computationally efficient, and the average running time for docking a peptide is ∼15 min with the range from about 1 min for short peptides to around 40 min for long peptides.

Combining Rosetta with molecular dynamics (MD): A benchmark of the MD-based ensemble protein design.

Computational protein design is a set of procedures for computing amino acid sequences that will fold into a specified structure. Rosetta Design, a commonly used software for protein design, allows for the effective identification of sequences compatible with a given backbone structure, while molecular dynamics (MD) simulations can thoroughly sample near-native conformations. We benchmarked a procedure in which Rosetta design is started on MD-derived structural ensembles and showed that such a combined approach generates 20-30% more diverse sequences than currently available methods with only a slight increase in computation time. Importantly, the increase in diversity is achieved without a loss in the quality of the designed sequences assessed by their resemblance to natural sequences. We demonstrate that the MD-based procedure is also applicable to de novo design tasks started from backbone structures without any sequence information. In addition, we implemented a protocol that can be used to assess the stability of designed models and to select the best candidates for experimental validation. In sum our results demonstrate that the MD ensemble-based flexible backbone design can be a viable method for protein design, especially for tasks that require a large pool of diverse sequences.

A novel knowledge-based potential for RNA 3D structure evaluation**Project supported by the National Science Foundation of China (Grants Nos. 11605125, 11105054, 11274124, and 11401448).

Ribonucleic acids (RNAs) play a vital role in biology, and knowledge of their three-dimensional (3D) structure is required to understand their biological functions. Recently structural prediction methods have been developed to address this issue, but a series of RNA 3D structures are generally predicted by most existing methods. Therefore, the evaluation of the predicted structures is generally indispensable. Although several methods have been proposed to assess RNA 3D structures, the existing methods are not precise enough. In this work, a new all-atom knowledge-based potential is developed for more accurately evaluating RNA 3D structures. The potential not only includes local and nonlocal interactions but also fully considers the specificity of each RNA by introducing a retraining mechanism. Based on extensive test sets generated from independent methods, the proposed potential correctly distinguished the native state and ranked near-native conformations to effectively select the best. Furthermore, the proposed potential precisely captured RNA structural features such as base-stacking and base-pairing. Comparisons with existing potential methods show that the proposed potential is very reliable and accurate in RNA 3D structure evaluation.