Search

Many scientific disciplines rely on computational methods for data analysis, model generation, and prediction. Implementing these methods is often accomplished by researchers with domain expertise but without formal training in software engineering or computer science. This arrangement has led to underappreciation of sustainability and maintainability of scientific software tools developed in academic environments. Some software tools have avoided this fate, including the scientific library Rosetta. We use this software and its community as a case study to show how modern software development can be accomplished successfully, irrespective of subject area. Rosetta is one of the largest software suites for macromolecular modeling, with 3.1 million lines of code and many state-of-the-art applications. Since the mid 1990s, the software has been developed collaboratively by the RosettaCommons, a community of academics from over 60 institutions worldwide with diverse backgrounds including chemistry, biology, physiology, physics, engineering, mathematics, and computer science. Developing this software suite has provided us with more than two decades of experience in how to effectively develop advanced scientific software in a global community with hundreds of contributors. Here we illustrate the functioning of this development community by addressing technical aspects (like version control, testing, and maintenance), community-building strategies, diversity efforts, software dissemination, and user support. We demonstrate how modern computational research can thrive in a distributed collaborative community. The practices described here are independent of subject area and can be readily adopted by other software development communities.

The knowledge of the three-dimensional structure of proteins is crucial for understanding many important biological processes. Most of the biologically relevant protein systems are too large for classical, atomistic molecular modeling tools. In such cases, coarse-grained (CG) models offer various opportunities for efficient conformational sampling and thus prediction of the three-dimensional structure. A variety of CG models have been proposed, each based on a similar framework consisting of a set of conceptual components such as protein representation, force field, sampling, etc. In this chapter we discuss these components, highlighting ideas which have proven to be the most successful. As CG methods are usually part of multistage procedures, we also describe approaches used for the incorporation of homology data and all-atom reconstruction methods.

We recently developed a new coarse-grained model of protein structure and dynamics [ Dawid et al. J. Chem. Theory Comput. 2017 , 13 ( 11 ), 5766 - 5779 ]. The model assumed a single bead representation of amino acid residues, where positions of such united residues were defined by centers of mass of four amino acid fragments. Replica exchange Monte Carlo sampling of the model chains provided good pictures of modeled structures and their dynamics. In its generic form the statistical knowledge-based force field of the model has been dedicated for single-domain globular proteins. Sequence-specific interactions are defined by three-letter secondary structure data. In the present work we demonstrate that different assignments and/or predictions of secondary structures are usually sufficient for enforcing cooperative formation of native-like folds of SURPASS chains for the majority of single-domain globular proteins. Simulations of native-like structure assembly for a representative set of globular proteins have shown that the accuracy of secondary structure data is usually not crucial for model performance, although some specific errors can strongly distort the obtained three-dimensional structures.

Coarse-grained modeling of biomolecules has a very important role in molecular biology. In this work we present a novel SURPASS (Single United Residue per Pre-Averaged Secondary Structure fragment) model of proteins that can be an interesting alternative for existing coarse-grained models. The design of the model is unique and strongly supported by the statistical analysis of structural regularities characteristic for protein systems. Coarse-graining of protein chain structures assumes a single center of interactions per residue and accounts for preaveraged effects of four adjacent residue fragments. Knowledge-based statistical potentials encode complex interaction patterns of these fragments. Using the Replica Exchange Monte Carlo sampling scheme and a generic version of the SURPASS force field we performed test simulations of a representative set of single-domain globular proteins. The method samples a significant part of conformational space and reproduces protein structures, including native-like, with surprisingly good accuracy. Future extension of the SURPASS model on large biomacromolecular systems is briefly discussed.

The traditional computational modeling of protein structure, dynamics, and interactions remains difficult for many protein systems. It is mostly due to the size of protein conformational spaces and required simulation time scales that are still too large to be studied in atomistic detail. Lowering the level of protein representation from all-atom to coarse-grained opens up new possibilities for studying protein systems. In this review we provide an overview of coarse-grained models focusing on their design, including choices of representation, models of energy functions, sampling of conformational space, and applications in the modeling of protein structure, dynamics, and interactions. A more detailed description is given for applications of coarse-grained models suitable for efficient combinations with all-atom simulations in multiscale modeling strategies.

The explosion of the size of the universe of known protein sequences has stimulated two complementary approaches to structural mapping of these sequences: theoretical structure prediction and experimental determination by structural genomics (SG). In this work, we assess the accuracy of structure prediction by two automated template-based structure prediction metaservers (genesilico.pl and bioinfo.pl) by measuring the structural similarity of the predicted models to corresponding experimental models determined a posteriori. Of 199 targets chosen from SG programs, the metaservers predicted the structures of about a fourth of them "correctly." (In this case, "correct" was defined as placing more than 70 % of the alpha carbon atoms in the model within 2 Å of the experimentally determined positions.) Almost all of the targets that could be modeled to this accuracy were those with an available template in the Protein Data Bank (PDB) with more than 25 % sequence identity. The majority of those SG targets with lower sequence identity to structures in the PDB were not predicted by the metaservers with this accuracy. We also compared metaserver results to CASP8 results, finding that the models obtained by participants in the CASP competition were significantly better than those produced by the metaservers.

Protein structure predictions, and experimentally derived protein structures, very often require certain structure improvement (refinement), which means bringing it closer to real, usually in vivo working conformations. In respect to the variety of protein models to be refined, computational optimization procedures could be divided into localized (applied to a small part of a structure) and global (whole structure). Generally speaking, the first problem is usually tractable, and the latter remains to be extremely challenging for systems larger then peptides or small proteins: optimization complexity and difficulty dramatically increase with the size of structures to be optimized. © 2012 John Wiley & Sons, Ltd.

Motivation: Knowledge-based potentials are valuable tools for protein structure modeling and evaluation of the quality of the structure prediction obtained by a variety of methods. Potentials of such type could be significantly enhanced by a proper exploitation of the evolutionary information encoded in related protein sequences. The new potentials could be valuable components of threading algorithms, ab-initio protein structure prediction, comparative modeling and structure modeling based on fragmentary experimental data. Results: A new potential for scoring local protein geometry is designed and evaluated. The approach is based on the similarity of short protein fragments measured by an alignment of their sequence profiles. Sequence specificity of the resulting energy function has been compared with the specificity of simpler potentials using gapless threading and the ability to predict specific geometry of protein fragments. Significant improvement in threading sensitivity and in the ability to generate sequence-specific protein-like conformations has been achieved. Availability: see: http://www.biocomp.chem.uw.edu.pl Contact: dgront@chem.uw.edu.pl

The knowledge of the three-dimensional structure of proteins is crucial for understanding many important biological processes. Most biologically important proteins are too large to handle for the classical simulation tools. In such cases, coarse-grained (CG) models nowadays offer various opportunities for efficient conformational sampling and thus prediction of the three-dimensional structure. A variety of CG models have been proposed, each based on a similar framework consisting of a set of conceptual components such as protein representation, force field, sampling, etc. In this chapter we discuss these components, highlighting ideas which have proven to be the most successful. As CG methods are usually part of multistage procedures, we also describe approaches used for the incorporation of homology data and all-atom reconstruction methods.

The problem of theoretical de novo protein structure prediction has been already investigated for a few decades. Throughout those years numerous different algorithms have been proposed to solve this problem. The most successful ones are capable of predicting the structure for small-size globular proteins (up to 80 amino acids). Recent years have also witnessed improvement in experimental structure determination methods, which became throughput and highly automated. Several steps however still have to be done manually. Combination of de novo prediction methods with fragmentary experimental data can be used to alleviate some of these bottlenecks.In our work we combined one of the most successful approaches for protein structure modeling: fragment recombination (ROSETTA method1) into a single protocol that employs fragmentary NMR data: Chemical Shifts, J-couplings as well as TEDOR and VEANS obtained in Solid State NMR experiments. The protocol, managed by BioShell2,3 software is very general and can utilize a wide variety of other than NMR types of data. The experimental restraints are applied on various stages of the procedure: to derive distance restraints, to select matching protein fragments from a structural database, to guide the conformational search and finally to score the obtained models. Our results show that in several cases it is possible to calculate a high-resolution protein structure without long-range experimental restraints i.e. solely from local backbone information. Moreover, the use of high-resolution lattice model greatly improves computational efficiency of the whole protocol.References1. R. Das and D. Baker, Ann. Rev. Biochem.77, 363-382 (2008).2. D. Gront and A. Kolinski, Bioinformatics22, 621-622 (2006).3. D. Gront and A. Kolinski, Bioinformatics 24, 584-585 (2008).

The structure of AF2331, a 11-kDa orphan protein of unknown function from Archaeoglobus fulgidus, was solved by Se-Met MAD to 2.4 A resolution. The structure consists of an alpha + beta fold formed by an unusual homodimer, where the two core beta-sheets are interdigitated, containing strands alternating from both subunits. The decrease in solvent-accessible surface area upon dimerization is unusually large (3960 A(2)) for a protein of its size. The percentage of the total surface area buried in the interface (41.1%) is one of the largest observed in a nonredundant set of homodimers in the PDB and is above the mean for nearly all other types of homo-oligomers. AF2331 has no sequence homologs, and no structure similar to AF2331 could be found in the PDB using the CE, TM-align, DALI, or SSM packages. The protein has been identified in Pfam 23.0 as the archetype of a new superfamily and is topologically dissimilar to all other proteins with the "3-Layer (BBA) Sandwich" fold in CATH. Therefore, we propose that AF2331 forms a novel alpha + beta fold. AF2331 contains multiple negatively charged surface clusters and is located on the same operon as the basic protein AF2330. We hypothesize that AF2331 and AF2330 may form a charge-stabilized complex in vivo, though the role of the negatively charged surface clusters is not clear.

We have developed and combined several novel methods to improve protein structure prediction from the amino acid sequence, and the structural refinement of protein models. One of the most promising developments in protein structure prediction are many-body potentials that take into account dense packing, and cooperativity of interactions in protein cores. We developed a method that uses whole protein information filtered through machine learners to score protein models based on their likeness to native structures. Testing on CASP 9 targets showed that our method is superior to the common DFIRE and its derivatives as well as to the current version of RWPlus, both of which are considered a standard in the field. By combing statistical contact potentials with entropies from the elastic network models of proteins we can compute free energy and improve coarse-grained modeling of protein structure and dynamics. The consideration of protein flexibility and its fluctuational dynamics improves protein structure prediction, and leads to a better refinement of computational models of proteins. We proposed a novel protein structural refinement procedure based on Anisotropic Network Model (ANM) of protein fluctuational dynamics and Go-like model of energy score. The starting structures were models from past CASP experiments. We changed positions of C-alpha atoms using ANM, creating a new set of 250 structures from the initial model, and computed energies of these structures using Go-like energy score. The top 5 coarse-grained structures were fully rebuilt with BBQ and Scrwl4. To remove bond stretches and the excluded volume clashes, short Molecular Mechanics simulations (up to 10,000 steps) were performed with OPLS-AA force field and implicit solvent GBSA-OBC. The whole structural refinement process was performed iteratively leading to the improvement of average RMSD from 3.8A to 2.6A in 50 iterations.

Conformations of globular proteins in the denatured state were studied using a high-resolution lattice model of proteins and Monte Carlo dynamics. The model assumes a united-atom and high-coordination lattice representation of the polypeptide conformational space. The force field of the model mimics the short-range protein-like conformational stiffness, hydrophobic interactions of the side chains and the main-chain hydrogen bonds. Two types of approximations for the short-range interactions were compared: simple statistical potentials and knowledge-based protein-specific potentials derived from the sequence-structure compatibility of short fragments of protein chains. Model proteins in the denatured state are relatively compact, although the majority of the sampled conformations are globally different from the native fold. At the same time short protein fragments are mostly native-like. Thus, the denatured state of the model proteins has several features of the molten globule state observed experimentally. Statistical potentials induce native-like conformational propensities in the denatured state, especially for the fragments located in the core of folded proteins. Knowledge-based protein-specific potentials increase only slightly the level of similarity to the native conformations, in spite of their qualitatively higher specificity in the native structures. For a few cases, where fairly accurate experimental data exist, the simulation results are in semiquantitative agreement with the physical picture revealed by the experiments. This shows that the model studied in this work could be used efficiently in computational studies of protein dynamics in the denatured state, and consequently for studies of protein folding pathways, i.e. not only for the modeling of folded structures, as it was shown in previous studies. The results of the present studies also provide a new insight into the explanation of the Levinthal's paradox.

Background The comparative modeling approach to protein structure prediction inherently relies on a template structure. Before building a model such a template protein has to be found and aligned with the query sequence. Any error made on this stage may dramatically affects the quality of result. There is a need, therefore, to develop accurate and sensitive alignment protocols. Results BioShell threading software is a versatile tool for aligning protein structures, protein sequences or sequence profiles and query sequences to a template structures. The software is also capable of sub-optimal alignment generation. It can be executed as an application from the UNIX command line, or as a set of Java classes called from a script or a Java application. The implemented Monte Carlo search engine greatly facilitates the development and benchmarking of new alignment scoring schemes even when the functions exhibit non-deterministic polynomial-time complexity. Conclusions Numerical experiments indicate that the new threading application offers template detection abilities and provides much better alignments than other methods. The package along with documentation and examples is available at: http://bioshell.pl/threading3d.

Background The majority of experimentally determined crystal structures of Type II restriction endonucleases (REases) exhibit a common PD-(D/E)XK fold. Crystal structures have been also determined for single representatives of two other folds: PLD (R.BfiI) and half-pipe (R.PabI), and bioinformatics analyses supported by mutagenesis suggested that some REases belong to the HNH fold. Our previous bioinformatic analysis suggested that REase R.Eco29kI shares sequence similarities with one more unrelated nuclease superfamily, GIY-YIG, however so far no experimental data were available to support this prediction. The determination of a crystal structure of the GIY-YIG domain of homing endonuclease I-TevI provided a template for modeling of R.Eco29kI and prompted us to validate the model experimentally. Results Using protein fold-recognition methods we generated a new alignment between R.Eco29kI and I-TevI, which suggested a reassignment of one of the putative catalytic residues. A theoretical model of R.Eco29kI was constructed to illustrate its predicted three-dimensional fold and organization of the active site, comprising amino acid residues Y49, Y76, R104, H108, E142, and N154. A series of mutants was constructed to generate amino acid substitutions of selected residues (Y49A, R104A, H108F, E142A and N154L) and the mutant proteins were examined for their ability to bind the DNA containing the Eco29kI site 5'-CCGCGG-3' and to catalyze the cleavage reaction. Experimental data reveal that residues Y49, R104, E142, H108, and N154 are important for the nuclease activity of R.Eco29kI, while H108 and N154 are also important for specific DNA binding by this enzyme. Conclusion Substitutions of residues Y49, R104, H108, E142 and N154 predicted by the model to be a part of the active site lead to mutant proteins with strong defects in the REase activity. These results are in very good agreement with the structural model presented in this work and with our prediction that R.Eco29kI belongs to the GIY-YIG superfamily of nucleases. Our study provides the first experimental evidence for a Type IIP REase that does not belong to the PD-(D/E)XK or HNH superfamilies of nucleases, and is instead a member of the unrelated GIY-YIG superfamily.

Background Although experimental methods for determining protein structure are providing high resolution structures, they cannot keep the pace at which amino acid sequences are resolved on the scale of entire genomes. For a considerable fraction of proteins whose structures will not be determined experimentally, computational methods can provide valuable information. The value of structural models in biological research depends critically on their quality. Development of high-accuracy computational methods that reliably generate near-experimental quality structural models is an important, unsolved problem in the protein structure modeling. Results Large sets of structural decoys have been generated using reduced conformational space protein modeling tool CABS. Subsequently, the reduced models were subject to all-atom reconstruction. Then, the resulting detailed models were energy-minimized using state-of-the-art all-atom force field, assuming fixed positions of the alpha carbons. It has been shown that a very short minimization leads to the proper ranking of the quality of the models (distance from the native structure), when the all-atom energy is used as the ranking criterion. Additionally, we performed test on medium and low accuracy decoys built via classical methods of comparative modeling. The test placed our model evaluation procedure among the state-of-the-art protein model assessment methods. Conclusion These test computations show that a large scale high resolution protein structure prediction is possible, not only for small but also for large protein domains, and that it should be based on a hierarchical approach to the modeling protocol. We employed Molecular Mechanics with fixed alpha carbons to rank-order the all-atom models built on the scaffolds of the reduced models. Our tests show that a physic-based approach, usually considered computationally too demanding for large-scale applications, can be effectively used in such studies.