QRNAS: software tool for refinement of nucleic acid structures

RNA molecules play fundamental roles in cellular processes. Their function and interactions with other biomolecules are dependent on the ability to form complex three-dimensional (3D) structures. However, experimental determination of RNA 3D structures is laborious and challenging, and therefore, the majority of known RNAs remain structurally uncharacterized. Here, we present SimRNA: a new method for computational RNA 3D structure prediction, which uses a coarse-grained representation, relies on the Monte Carlo method for sampling the conformational space, and employs a statistical potential to approximate the energy and identify conformations that correspond to biologically relevant structures. SimRNA can fold RNA molecules using only sequence information, and, on established test sequences, it recapitulates secondary structure with high accuracy, including correct prediction of pseudoknots. For modeling of complex 3D structures, it can use additional restraints, derived from experimental or computational analyses, including information about secondary structure and/or long-range contacts. SimRNA also can be used to analyze conformational landscapes and identify potential alternative structures.

In analogy to proteins, the function of RNA depends on its structure and dynamics, which are encoded in the linear sequence. While there are numerous methods for computational prediction of protein 3D structure from sequence, there have been however very few such methods for RNA. This chapter discusses template-based and template-free approaches for macromolecular structure prediction, with special emphasis on comparison between the already tried-and-tested methods for protein structure modeling and the very recently developed “protein-like” modeling methods for RNA. As examples, we briefly review our recently developed tools for RNA 3D structure prediction, including ModeRNA (template-based or comparative/homology modeling) and SimRNA (template-free or de novo modeling). ModeRNA requires, as an input, atomic 3D coordinates of a template RNA molecule and a user-specified sequence alignment between the target to be modeled and the template. It can model posttranscriptional modifications, a functionally important feature analogous to posttranslational modifications in proteins. It can model the structures of RNAs of essentially any length, provided that a starting template is known. SimRNA can fold RNA 3D structure starting from sequence alone. It is based on a coarse-grained representation of the polynucleotide chains (only three atoms per nucleotide) and uses a Monte Carlo sampling scheme to generate moves in the 3D space, with a statistical potential to estimate the free energy. The current implementation based on simulated annealing is able to find native-like conformations for RNAs <100 nt in length, with multiple runs required to fold long sequences.

In analogy to proteins, the function of RNA depends on its structure and dynamics, which are encoded in the linear sequence. While there are numerous methods for computational prediction of protein 3D structure from sequence, there have been very few such methods for RNA. This review discusses template-based and template-free approaches for macromolecular structure prediction, with special emphasis on comparison between the already tried-and-tested methods for protein structure modeling and the very recently developed “protein-like” modeling methods for RNA. We highlight analogies between many successful methods for modeling of these two types of biological macromolecules and argue that RNA 3D structure can be modeled using “protein-like” methodology. We also highlight the areas where the differences between RNA and proteins require the development of RNA-specific solutions.FigureApproaches for predicting RNA structure. Top: Template-free modeling. Bottom: Template-based modeling

The goal of protein structure prediction is to estimate the spatial position of every atom of protein molecules from the amino acid sequence by computational methods. Depending on the availability of homologous templates in the PDB library, structure prediction approaches are categorised into template‐based modelling (TBM) and free modelling (FM). While TBM is by far the only reliable method for high‐resolution structure prediction, challenges in the field include constructing the correct folds without using template structures and refining the template models closer to the native state when templates are available. Nevertheless, the usefulness of various levels of protein structure predictions have been convincingly demonstrated in biological and medical applications. Key Concepts: Evolution is a general principle to guide protein structure and function predictions. Proteins of similar sequence have similar 3D structure. Function of protein is decided by the 3D structure. TBM using homologous templates has the highest accuracy. Template structure can be refined by combining multiple templates. Current physics‐based ab initio folding can only fold small proteins. Threading is an efficient tool for detecting distantly homologous templates. Membrane protein structure prediction is challenging due to the lack of templates. Disordered regions exist in protein which does not possess stable structure but has important function implications.

Since Anfinsen demonstrated that the information encoded in a protein’s amino acid sequence determines its structure in 1973, solving the protein structure prediction problem has been the Holy Grail of structural biology. The goal of protein structure prediction approaches is to utilize computational modeling to determine the spatial location of every atom in a protein molecule starting from only its amino acid sequence. Depending on whether homologous structures can be found in the Protein Data Bank (PDB), structure prediction methods have been historically categorized as template-based modeling (TBM) or template-free modeling (FM) approaches. Until recently, TBM has been the most reliable approach to predicting protein structures, and in the absence of reliable templates, the modeling accuracy sharply declines. Nevertheless, the results of the most recent community-wide assessment of protein structure prediction experiment (CASP14) have demonstrated that the protein structure prediction problem can be largely solved through the use of end-to-end deep machine learning techniques, where correct folds could be built for nearly all single-domain proteins without using the PDB templates. Critically, the model quality exhibited little correlation with the quality of available template structures, as well as the number of sequence homologs detected for a given target protein. Thus, the implementation of deep-learning techniques has essentially broken through the 50-year-old modeling border between TBM and FM approaches and has made the success of high-resolution structure prediction significantly less dependent on template availability in the PDB library.

G-protein-coupled receptors are of great pharmaceutical interest, comprising the majority of targets for currently marketed drugs. The theme of my thesis is the development of the structure prediction method, MembStruk, for the superfamily of G-protein-coupled receptors. The first part of this thesis focuses on the methods and their validation. There are several steps involved in MembStruk that are detailed and tested for membrane proteins with known structures in the first few chapters (Chapters 2-6). Specifically, the first principles methods for predicting the transmembrane helical ranges and the helix hydrophobic centers are tested. The program for predicting the transmembrane helical ranges, TM2ndS, ranks in the top two when comparing performance with other top prediction methods. And because it is based on general principles, it can be applied robustly for membrane protein families for which little structural information is available. The simulation of the EC-II closing is also tested on bovine rhodopsin. The use of the MembStruk method on bovine rhodopsin as a validation case is presented in detail (Chapter 2). The large majority (71%) of the residues involved in binding in rhodopsin are predicted and the protein structure itself is 2.84 [...] coordinate root mean square error in the transmembrane main chain atoms from the crystal structure. The second part of the thesis discusses applications on various G-protein-coupled receptor systems. The application of the MembStruk method to other peptide chemokine G-protein-coupled receptors like CCR1 and CCR5 is discussed in Chapter 9. The fundamental scientific problems of G-protein-coupled receptor modulation of absorption and relaxation properties of a bound chromophore (retinal) are addressed and results are presented for the predictions of these properties. The prediction of structure and function of G-protein-coupled receptors would allow for structure-based drug design and a rational approach to reducing drug cross-reactivity across receptor families.

G-Protein Coupled Receptors (GPCRs) form a major target class of membrane proteins for therapeutic drug design, and the challenge is to design subtype specific drugs. Hence the knowledge of three-dimensional structure is critical to drug design for GPCRs. Since GPCRs are membrane bound proteins, there is only one crystal structure for a GPCR, namely bovine rhodopsin. The prediction of structure and function of G-protein-coupled receptors will allow for designing drugs with minimal side effects. The focus of my thesis is the development of computational methods for prediction of structure of GPCRs and application of these methods (MembStruk) for a class of important drug targets such as chemokine receptors. MembStruk method is a hierarchical method ranging from coarse grain optimization of the trans-membrane helices to fine grain optimization of the structure in explicit lipid bilayer. The first two chapters of the thesis details the computational steps involved in MembStruk and its application to validating the method for bovine rhodopsin. The first chapter presents the method development in the most current version of the MembStruk method, version 4.30, and its application to bovine rhodopsin. The final predicted structure for bovine rhodopsin deviates from the crystal structure trans-membrane main chain atoms by 2.66 A coordinate root mean square deviation (CRMSD), and the residues in the binding site of 11cis-retinal is only 1.37 A CRMSD from the crystal structure for the main chain atoms. The second chapter of this thesis details the computational methods for optimization of the rotation and translation of the trans-membrane regions. These methods of rotation and translation of transmembrane helices has been further extended to the comparison of structures of two membrane proteins, and applied to the comparison of crystal structures of bovine rhodopsin and bacteriorhodopsin. The third chapter details the graphical user interface that has been developed to automate the various steps of the MembStruk method. Olfactory receptors are GPCRs and the molecular analysis for the recognition of odorants is very important in understanding the mechanism of olfaction. In a blind study prior to experiments, in collaboration with Dr. Bozza of Rockefeller University, I applied the MembStruk method to understanding the binding of odorants to rat and mouse olfactory receptor I7. Chapter 4 describes the application of the MembStruk method to rat and mouse I7 olfactory receptor and the binding of 65 odorants to this receptor. The last chapter describes the use of MembStruk method in predicting the structure and function of important drug targets, namely chemokine receptors CCR5 and CXCR4.

Atomic-Level Protein Structure Refinement Using Fragment-Guided Molecular Dynamics Conformation Sampling

This chapter gives an overview over the current methods for automated modeling of RNA structures, with emphasis on template-based methods. The currently used approaches to RNA modeling are presented with a side view on the protein world, where many similar ideas have been used. Two main programs for automated template-based modeling are presented: ModeRNA assembling structures from fragments and MacroMoleculeBuilder performing a simulation to satisfy spatial restraints. Both approaches have in common that they require an alignment of the target sequence to a known RNA structure that is used as a modeling template. As a way to find promising template structures and to align the target and template sequences, we propose a pipeline combining the ParAlign and Infernal programs on RNA family data from Rfam. We also briefly summarize template-free methods for RNA 3D structure prediction. Typically, RNA structures generated by automated modeling methods require local or global optimization. Thus, we also discuss methods that can be used for local or global refinement of RNA structures.

Protein comparative modeling has useful applications in large-scale structural initiatives and in rational design of drug targets in medicinal chemistry. The reliability of a homology model is dependent on the sequence identity between the query and the structural homologue used as a template for modeling. Here, we present a method for the utilization and conservation of important structural features of template structures by providing additional spatial restraints in comparative modeling programs like MODELLER. We show that root mean square deviation at C(alpha) positions between the model and the corresponding experimental structure and the quality of the models can be significantly improved for distantly related systems by utilizing additional spatial restraints of the template structures. We demonstrate the influence of such approaches to homology modeling during distant relationships in understanding functional properties of protein such as ligand binding using cytochrome P450 as an example.

Protein structure prediction is an important problem in bioinformatics and has been studied for decades. However, there are still few open-source comprehensive protein structure prediction packages publicly available in the field. In this paper, we present our latest open-source protein tertiary structure prediction system—MULTICOM2, an integration of template-based modeling (TBM) and template-free modeling (FM) methods. The template-based modeling uses sequence alignment tools with deep multiple sequence alignments to search for structural templates, which are much faster and more accurate than MULTICOM1. The template-free (ab initio or de novo) modeling uses the inter-residue distances predicted by DeepDist to reconstruct tertiary structure models without using any known structure as template. In the blind CASP14 experiment, the average TM-score of the models predicted by our server predictor based on the MULTICOM2 system is 0.720 for 58 TBM (regular) domains and 0.514 for 38 FM and FM/TBM (hard) domains, indicating that MULTICOM2 is capable of predicting good tertiary structures across the board. It can predict the correct fold for 76 CASP14 domains (95% regular domains and 55% hard domains) if only one prediction is made for a domain. The success rate is increased to 3% for both regular and hard domains if five predictions are made per domain. Moreover, the prediction accuracy of the pure template-free structure modeling method on both TBM and FM targets is very close to the combination of template-based and template-free modeling methods. This demonstrates that the distance-based template-free modeling method powered by deep learning can largely replace the traditional template-based modeling method even on TBM targets that TBM methods used to dominate and therefore provides a uniform structure modeling approach to any protein. Finally, on the 38 CASP14 FM and FM/TBM hard domains, MULTICOM2 server predictors (MULTICOM-HYBRID, MULTICOM-DEEP, MULTICOM-DIST) were ranked among the top 20 automated server predictors in the CASP14 experiment. After combining multiple predictors from the same research group as one entry, MULTICOM-HYBRID was ranked no. 5. The source code of MULTICOM2 is freely available at https://github.com/multicom-toolbox/multicom/tree/multicom_v2.0.

Protein Structure Prediction From Primary Sequence

RNA encompasses an essential part of all known forms of life. The functions of many RNA molecules are dependent on their ability to form complex three-dimensional (3D) structures. However, experimental determination of RNA 3D structures is laborious and challenging, and therefore, the majority of known RNAs remain structurally uncharacterized. To address this problem, computational structure prediction methods were developed that either utilize information derived from known structures of other RNA molecules (by way of template-based modeling) or attempt to simulate the physical process of RNA structure formation (by way of template-free modeling). All computational methods suffer from various limitations that make theoretical models less reliable than high-resolution experimentally determined structures. This chapter provides a protocol for computational modeling of RNA 3D structure that overcomes major limitations by combining two complementary approaches: template-based modeling that is capable of predicting global architectures based on similarity to other molecules but often fails to predict local unique features, and template-free modeling that can predict the local folding, but is limited to modeling the structure of relatively small molecules. Here, we combine the use of a template-based method ModeRNA with a template-free method SimRNA. ModeRNA requires a sequence alignment of the target RNA sequence to be modeled with a template of the known structure; it generates a model that predicts the structure of a conserved core and provides a starting point for modeling of variable regions. SimRNA can be used to fold small RNAs (<80 nt) without any additional structural information, and to refold parts of models for larger RNAs that have a correctly modeled core. ModeRNA can be either downloaded, compiled and run locally or run through a web interface at http://genesilico.pl/modernaserver/ . SimRNA is currently available to download for local use as a precompiled software package at http://genesilico.pl/software/stand-alone/simrna and as a web server at http://genesilico.pl/SimRNAweb . For model optimization we use QRNAS, available at http://genesilico.pl/qrnas .

RNA 3D structures are critical for understanding their functions. However, only a limited number of RNA structures have been experimentally solved, so computational prediction methods are highly desirable. Nevertheless, accurate prediction of RNA 3D structures, especially those containing multiway junctions, remains a significant challenge, mainly due to the complicated non-canonical base pairing and stacking interactions in the junction loops and the possible long-range interactions between loop structures. Here we present RNAJP (‘RNA Junction Prediction’), a nucleotide- and helix-level coarse-grained model for the prediction of RNA 3D structures, particularly junction structures, from a given 2D structure. Through global sampling of the 3D arrangements of the helices in junctions using molecular dynamics simulations and in explicit consideration of non-canonical base pairing and base stacking interactions as well as long-range loop–loop interactions, the model can provide significantly improved predictions for multibranched junction structures than existing methods. Moreover, integrated with additional restraints from experiments, such as junction topology and long-range interactions, the model may serve as a useful structure generator for various applications.

G protein-coupled receptors (GPCRs) represent a major focus in functional genomics programs and drug development research, but their important potential as drug targets contrasts with the still limited data available concerning their activation mechanism. Here, we investigated the activation mechanism of the cholecystokinin-2 receptor (CCK2R). The three-dimensional structure of inactive CCK2R was homology-modeled on the basis of crystal coordinates of inactive rhodopsin. Starting from the inactive CCK2R modeled structure, active CCK2R (namely cholecystokinin-occupied CCK2R) was modeled by means of steered molecular dynamics in a lipid bilayer and by using available data from other GPCRs, including rhodopsin. By comparing the modeled structures of the inactive and active CCK2R, we identified changes in the relative position of helices and networks of interacting residues, which were expected to stabilize either the active or inactive states of CCK2R. Using targeted molecular dynamics simulations capable of converting CCK2R from the inactive to the active state, we delineated structural changes at the atomic level. The activation mechanism involved significant movements of helices VI and V, a slight movement of helices IV and VII, and changes in the position of critical residues within or near the binding site. The mutation of key amino acids yielded inactive or constitutively active CCK2R mutants, supporting this proposed mechanism. Such progress in the refinement of the CCK2R binding site structure and in knowledge of CCK2R activation mechanisms will enable target-based optimization of nonpeptide ligands.