Residue coevolution estimations coupled to machine learning methods are revolutionizing the ability of protein structure prediction approaches to model proteins that lack clear homologous templates in the Protein Data Bank (PDB). This has been patent in the last round of the Critical Assessment of Structure Prediction (CASP), which presented several very good models for the hardest targets. Unfortunately, literature reporting on these advances often lacks digests tailored to lay end users; moreover, some of the top-ranking predictors do not provide webservers that can be used by nonexperts. How can then end users benefit from these advances and correctly interpret the predicted models? Here we review the web resources that biologists can use today to take advantage of these state-of-the-art methods in their research, including not only the best de novo modeling servers but also datasets of models precomputed by experts for structurally uncharacterized protein families. We highlight their features, advantages and pitfalls for predicting structures of proteins without clear templates. We present a broad number of applications that span from driving forward biochemical investigations that lack experimental structures to actually assisting experimental structure determination in X-ray diffraction, cryo-EM and other forms of integrative modeling. We also discuss issues that must be considered by users yet still require further developments, such as global and residue-wise model quality estimates and sources of residue coevolution other than monomeric tertiary structure.

MOTIVATION: Proteins are intrinsically dynamic entities. Flexibility sampling methods, such as molecular dynamics or those arising from integrative modeling strategies, are now commonplace and enable the study of molecular conformational landscapes in many contexts. Resulting structural ensembles increase in size as technological and algorithmic advancements take place, making their analysis increasingly demanding. In this regard, cluster analysis remains a go-to approach for their classification. However, many state-of-the-art algorithms are restricted to specific cluster properties. Combined with tedious parameter fine-tuning, cluster analysis of protein structural ensembles suffers from the lack of a generally applicable and easy to use clustering scheme. RESULTS: We present CLoNe, an original Python-based clustering scheme that builds on the Density Peaks algorithm of Rodriguez and Laio. CLoNe relies on a probabilistic analysis of local density distributions derived from nearest neighbors to find relevant clusters regardless of cluster shape, size, distribution and amount. We show its capabilities on many toy datasets with properties otherwise dividing state-of-the-art approaches and improves on the original algorithm in key aspects. Applied to structural ensembles, CLoNe was able to extract meaningful conformations from membrane binding events and ligand-binding pocket opening as well as identify dominant dimerization motifs or inter-domain organization. CLoNe additionally saves clusters as individual trajectories for further analysis and provides scripts for automated use with molecular visualization software. AVAILABILITY AND IMPLEMENTATION: www.epfl.ch/labs/lbm/resources, github.com/LBM-EPFL/CLoNe. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Modern research in life sciences relies vitally on experimental 3D structures and also, undeniably, on models derived through computational and integrative modeling methods. Unfortunately, while high-resolution 3D models from experimental structure determination are well centralized and made easily discoverable by the wwPDB, the same is far from true for computational and integrative models, which remain largely unavailable, difficult to find and disconnected from literature. On the one hand, deposition of such data upon publication is not mandatory, and on the other hand, many existing resources of molecular models are not easily searchable. The purpose of this letter is to call for (i) design and implementation of policies to make computational models openly available and (ii) institution of a definitive, centralized resource that facilitates reuse and discovery of computational models associated to literature.AQ4 Only ∼13% of PubMed articles that used structural models of proteins at the core of the work, sampled from publications of the last year, made the models freely available online (Supplementary Table S1). This is despite the numerous databases available where authors could seamlessly share their models: Model Archive (Schwede et al., 2009), Protein Model Data Bank (Castrignanò et al., 2006), Protein Model Portal (Haas et al., 2013), SwissModel’s and ModBase’s repositories for pure computational models (Pieper et al., 2014; Waterhouse et al., 2018); PDB-Dev (Berman et al., 2019; Burley et al., 2017; Sali et al., 2015) for integrative models; and SASBDB for SAXS-based models (Valentini et al., 2015). Some authors do not even use these services even though they are willing to share their models, which they do as Supplementary Material at publisher’s and their own websites (examples in Supplementary Table S1). We argue that models should be no exception to science-opening policies as depicted by the FAIR principles (Stall et al., 2019; Wilkinson et al., 2016). Models should be made available upon publication just like any other methodological or supporting data that is mandatory for the sakes of reproducibility, reuse, reanalysis and criticism. Our proposal could well apply not only to homology and ab initio models but also to multi-molecule complexes obtained by docking and virtual screening, and to relevant conformations sampled in molecular simulations. Notably, open availability of such data would help not only to judiciously follow the written descriptions from publications in interactive 3D, but would also serve as material for further studies, including future experimental structure determinations [as in Garcia-Alai et al. (2018)]. Open model availability would be especially valuable for models generated with expensive computer resources and very specific expertise. This includes tertiary structure models based on contact/distance predictions plus extensive conformational sampling (Greener et al., 2019; Michel et al., 2017; Ovchinnikov et al., 2015, 2017; Wang et al., 2019) as recent CASP rounds have highlighted (Abriata et al., 2018, 2019), poses from molecular dynamics simulations, which enable investigations unparalleled by experimentation alone (Bottaro and Lindorff-Larsen, 2018), and models from integrative methods that take the most out of experimental data (Rout and Sali, 2019; Vallat et al., 2019). Storage is probably not a limitation, especially with modern repositories like CERN-based Zenodo, which can even host simulation trajectories at reasonable stride (as many researchers are now starting to share). Since the technology and resources required to build the hub that molecular models deserve are easily available, we are only faced with the rather political challenge of reaching agreement among publishers, funding agencies, groups running modeling services and competitions (CASP, CAPRI, etc.) and the varied existing databases of literature-associated models, to set up a definitive, centralized hub that facilitates model deposition, annotation, discovery and reutilization. Such hub (Fig. 1) would allow the community to better benefit from the recent improvements in protein structure prediction, the rise in tools for integrative modeling, and the deep exploratory power of modern molecular simulations. Working of a centralizing archive for models of biological macromolecules, containing only sequences and links to the model-hosting servers. The central archive is a database of protein sequences associated to URLs or DOIs that point to the corresponding citations and downloadable models. This way there is no duplication of model coordinates, keeping a compact web server and allowing authors to submit model coordinates to the most appropriate database together with other relevant data. From the point of view of the users, besides allowing them to directly download models for inspection (green), such database also allows them to find models that match or resemble a query sequence (red), which has useful applications as exemplified in Figure 2. (Color version of this figure is available at Bioinformatics online.) Although one can of course get models from authors upon request, this is far less efficient than direct downloads and does not solve the problem of discoverability. This takes us to the second point of this letter. It would be useful that model-hosting servers allow finding models by routes that do not require knowing about the reporting articles; for example models of proteins should ideally be discoverable through sequence queries. Surprisingly, however, many modern databases and datasets of models, especially those based on methodologies other than comparative modeling (PDB-Dev, SASBDB, published datasets of contact/distance prediction-based models), do not allow sequence-level searches. Some of the available model archives like the Protein Model Portal (Haas et al., 2013), the Protein Model Data Base (Castrignanò et al., 2006) and the Model Archive (Schwede et al., 2009) have no sequence-search functionalities, or have them but unsupported (in particular, the Protein Model Portal will soon be discontinued). One notable resource that has a working sequence-search system is ModBase (Pieper et al., 2014), but this resource is limited to models derived from homology modeling. Today, the most interesting models, i.e. those for proteins that share no detectable similarity to any PDB entry thus being unamenable to homology modeling, those observed in docking and molecular simulation studies and those involving larger assemblies derived through integrative modeling methods, are all hardly discoverable. For all these reasons, we call for agreement in creating a single repository with features that facilitate discovery, especially through sequence queries. An important effort is ongoing but specifically tailored to integrative models and not yet incorporating sequence-search functionalities (Berman et al., 2019). There is however no technological impediment, actually the simplest solution would be convincing journals of mandating model deposition to one of the existing archives, or even to Zenodo, and then setting up a sequence-searchable database that links each sequence to its location (Fig. 1). We have in fact devised a palliative solution following this idea at http://lucianoabriata.altervista.org/modelsearch/ that allows querying for >5000 sequences for which models are available in resources based primarily on methods other than comparative modeling (examples in Fig. 2). More precisely, >2900 entries come from contact prediction-based modeling of proteins of unknown structures and no clear templates [from Gremlin/Rosetta (Ovchinnikov et al., 2015, 2017), PconsC3 (Michel et al., 2017), DMPfold (Greener et al., 2019) and Tara (Wang et al., 2019)], >1200 come from CASP models (most corresponding to experimental structures already available at the wwPDB, but some were never released hence their value), 250 entries come from integrative models at PDB-Dev, ∼700 entries come from SASBDB and ∼100 were manually entered from the Supplementary Material of the literature inspected for the statistics commented above and other resources. Models become much more useful if discoverable through sequence searches. Example (A) is a target from Salmonella typhimurium for which no templates are currently available in the Protein Data Bank and therefore no structural models in standard databases of homology models. However, the search identifies a model derived through contact predictions for a protein with similar sequence to the target’s C-terminal domain, enabling confident structure prediction for at least this domain using the identified model as template. In example (B), searching ubiquitin’s sequence retrieves models of covalent ubiquitin dimers and oligomers derived through integrative modeling, from the PDB-Dev and SASBDB Clearly, an actual service could not rely on a web app at its core but would rather require server-side computing to efficiently search for sequences among very large numbers of models. This would require funding to maintain the website and a host group that could possibly branch from an existing structural biology resource like the wwPDB or CASP. At least in a first stage to reduce costs, model storage could rely on third parties like the repositories mentioned above. To ensure reliability, only models from peer-reviewed publications should be accepted, which implies close coordination with publishers and libraries like Pubmed. From the user viewpoint, it is of utmost importance that entries also include quality estimates, both provided by the authors of the model and from automated methods (Won et al., 2019). Importantly, quality estimate methods need themselves to be evaluated and agreed upon. Other features like those of wwPDB sites, such as online visualization and connection to other databases, would also be possible in a dedicated website, and very useful. Academia, industry, funding agencies and scholarly publishers have come together to design and jointly endorse the FAIR Data Principles. These principles emphasize on enhancing the ability of machines to automatically find and use data, and supporting data reuse by individuals (Stall et al., 2019; Wilkinson et al., 2016). As others have also recently highlighted (Abraham et al., 2019; Graham et al., 2019; Vallat et al., 2019), it is now about time that these sharing principles be applied to structural models of biological molecules too, which will ultimately help in the endeavor of structurally annotating genomes (Sillitoe et al., 2020). We acknowledge all the colleagues with whom we discussed in the CASP13 meeting about the issues of opening up models and making the community of biologists aware about the utility of models for their research. Financial Support: none declared. Conflict of Interest: none declared.

BACKGROUND: Protein variability can now be studied by measuring high-resolution tolerance-to-substitution maps and fitness landscapes in saturated mutational libraries. But these rich and expensive datasets are typically interpreted coarsely, restricting detailed analyses to positions of extremely high or low variability or dubbed important beforehand based on existing knowledge about active sites, interaction surfaces, (de)stabilizing mutations, etc. RESULTS: Our new webserver PsychoProt (freely available without registration at http://psychoprot.epfl.ch or at http://lucianoabriata.altervista.org/psychoprot/index.html ) helps to detect, quantify, and sequence/structure map the biophysical and biochemical traits that shape amino acid preferences throughout a protein as determined by deep-sequencing of saturated mutational libraries or from large alignments of naturally occurring variants. DISCUSSION: We exemplify how PsychoProt helps to (i) unveil protein structure-function relationships from experiments and from alignments that are consistent with structures according to coevolution analysis, (ii) recall global information about structural and functional features and identify hitherto unknown constraints to variation in alignments, and (iii) point at different sources of variation among related experimental datasets or between experimental and alignment-based data. Remarkably, metabolic costs of the amino acids pose strong constraints to variability at protein surfaces in nature but not in the laboratory. This and other differences call for caution when extrapolating results from in vitro experiments to natural scenarios in, for example, studies of protein evolution. CONCLUSION: We show through examples how PsychoProt can be a useful tool for the broad communities of structural biology and molecular evolution, particularly for studies about protein modeling, evolution and design.

The PhoQP two-component system is a signaling complex essential for bacterial virulence and cationic antimicrobial peptide resistance. PhoQ is the histidine kinase chemoreceptor of this tandem machine and assembles in a homodimer conformation spanning the bacterial inner membrane. Currently, a full understanding of the PhoQ signal transduction is hindered by the lack of a complete atomistic structure. In this study, an atomistic model of the key transmembrane (TM) domain is assembled by using molecular simulations, guided by experimental cross-linking data. The formation of a polar pocket involving Asn202 in the lumen of the tetrameric TM bundle is crucial for the assembly and solvation of the domain. Moreover, a concerted displacement of the TM helices at the periplasmic side is found to modulate a rotation at the cytoplasmic end, supporting the transduction of the chemical signal through a combination of scissoring and rotational movement of the TM helices.