Combining machine learning with structure-based protein design to predict and engineer post-translational modifications of proteins.

Post-translational modifications (PTMs) of proteins play a vital role in their function and stability. These modifications influence protein folding, signaling, protein-protein interactions, enzyme activity, binding affinity, aggregation, degradation, and much more. To date, over 400 types of PTMs have been described, representing chemical diversity well beyond the genetically encoded amino acids. Such modifications pose a challenge to the successful design of proteins, but also represent a major opportunity to diversify the protein engineering toolbox. To this end, we first trained artificial neural networks (ANNs) to predict eighteen of the most abundant PTMs, including protein glycosylation, phosphorylation, methylation, and deamidation. In a second step, these models were implemented inside the computational protein modeling suite Rosetta, which allows flexible combination with existing protocols to model the modified sites and understand their impact on protein stability as well as function. Lastly, we developed a new design protocol that either maximizes or minimizes the predicted probability of a particular site being modified. We find that this combination of ANN prediction and structure-based design can enable the modification of existing, as well as the introduction of novel, PTMs. The potential applications of our work include, but are not limited to, glycan masking of epitopes, strengthening protein-protein interactions through phosphorylation, as well as protecting proteins from deamidation liabilities. These applications are especially important for the design of new protein therapeutics where PTMs can drastically change the therapeutic properties of a protein. Our work adds novel tools to Rosetta's protein engineering toolbox that allow for the rational design of PTMs.

Recent Advances in Automated Structure-Based De Novo Drug Design.

As the number of determined and predicted protein structures and the size of druglike 'make-on-demand' libraries soar, the time-consuming nature of structure-based computer-aided drug design calls for innovative computational algorithms. De novo drug design introduces in silico heuristics to accelerate searching in the vast chemical space. This review focuses on recent advances in structure-based de novo drug design, ranging from conventional fragment-based methods, evolutionary algorithms, and Metropolis Monte Carlo methods to deep generative models. Due to the historical limitation of de novo drug design generating readily available drug-like molecules, we highlight the synthetic accessibility efforts in each category and the benchmarking strategies taken to validate the proposed framework.

Computationally designed peptide macrocycle inhibitors of New Delhi metallo-ß-lactamase 1.

The rise of antibiotic resistance calls for new therapeutics targeting resistance factors such as the New Delhi metallo-ß-lactamase 1 (NDM-1), a bacterial enzyme that degrades ß-lactam antibiotics. We present structure-guided computational methods for designing peptide macrocycles built from mixtures of l- and d-amino acids that are able to bind to and inhibit targets of therapeutic interest. Our methods explicitly consider the propensity of a peptide to favor a binding-competent conformation, which we found to predict rank order of experimentally observed IC50 values across seven designed NDM-1- inhibiting peptides. We were able to determine X-ray crystal structures of three of the designed inhibitors in complex with NDM-1, and in all three the conformation of the peptide is very close to the computationally designed model. In two of the three structures, the binding mode with NDM-1 is also very similar to the design model, while in the third, we observed an alternative binding mode likely arising from internal symmetry in the shape of the design combined with flexibility of the target. Although challenges remain in robustly predicting target backbone changes, binding mode, and the effects of mutations on binding affinity, our methods for designing ordered, binding-competent macrocycles should have broad applicability to a wide range of therapeutic targets.

Modeling Immunity with Rosetta: Methods for Antibody and Antigen Design.

Structure-based antibody and antigen design has advanced greatly in recent years, due not only to the increasing availability of experimentally determined structures but also to improved computational methods for both prediction and design. Constant improvements in performance within the Rosetta software suite for biomolecular modeling have given rise to a greater breadth of structure prediction, including docking and design application cases for antibody and antigen modeling. Here, we present an overview of current protocols for antibody and antigen modeling using Rosetta and exemplify those by detailed tutorials originally developed for a Rosetta workshop at Vanderbilt University. These tutorials cover antibody structure prediction, docking, and design and antigen design strategies, including the addition of glycans in Rosetta. We expect that these materials will allow novice users to apply Rosetta in their own projects for modeling antibodies and antigens.

Better together: Elements of successful scientific software development in a distributed collaborative community.

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.

Identification of a ubiquitin-binding interface using Rosetta and DEER.

ExoU is a type III-secreted cytotoxin expressing A2 phospholipase activity when injected into eukaryotic target cells by the bacterium Pseudomonas aeruginosa The enzymatic activity of ExoU is undetectable in vitro unless ubiquitin, a required cofactor, is added to the reaction. The role of ubiquitin in facilitating ExoU enzymatic activity is poorly understood but of significance for designing inhibitors to prevent tissue injury during infections with strains of P. aeruginosa producing this toxin. Most ubiquitin-binding proteins, including ExoU, demonstrate a low (micromolar) affinity for monoubiquitin (monoUb). Additionally, ExoU is a large and dynamic protein, limiting the applicability of traditional structural techniques such as NMR and X-ray crystallography to define this protein-protein interaction. Recent advancements in computational methods, however, have allowed high-resolution protein modeling using sparse data. In this study, we combine double electron-electron resonance (DEER) spectroscopy and Rosetta modeling to identify potential binding interfaces of ExoU and monoUb. The lowest-energy scoring model was tested using biochemical, biophysical, and biological techniques. To verify the binding interface, Rosetta was used to design a panel of mutations to modulate binding, including one variant with enhanced binding affinity. Our analyses show the utility of computational modeling when combined with sensitive biological assays and biophysical approaches that are exquisitely suited for large dynamic proteins.

Engineering an allosteric transcription factor to respond to new ligands.

Genetic regulatory proteins inducible by small molecules are useful synthetic biology tools as sensors and switches. Bacterial allosteric transcription factors (aTFs) are a major class of regulatory proteins, but few aTFs have been redesigned to respond to new effectors beyond natural aTF-inducer pairs. Altering inducer specificity in these proteins is difficult because substitutions that affect inducer binding may also disrupt allostery. We engineered an aTF, the Escherichia coli lac repressor, LacI, to respond to one of four new inducer molecules: fucose, gentiobiose, lactitol and sucralose. Using computational protein design, single-residue saturation mutagenesis or random mutagenesis, along with multiplex assembly, we identified new variants comparable in specificity and induction to wild-type LacI with its inducer, isopropyl ß-D-1-thiogalactopyranoside (IPTG). The ability to create designer aTFs will enable applications including dynamic control of cell metabolism, cell biology and synthetic gene circuits.

Mono(2-ethylhexyl) phthalate (MEHP) and mono(2-ethyl-5-oxohexyl) phthalate (MEOHP) but not di(2-ethylhexyl) phthalate (DEHP) bind productively to the peroxisome proliferator-activated receptor γ.

RATIONALE: The most frequently occurring phthalate, di(2-ethylhexyl) phthalate (DEHP), causes adverse effects on glucose homeostasis and insulin sensitivity in several cell models and epidemiological studies. However, thus far, there is no information available on the molecular interaction of phthalates and one of the key regulators of the metabolism, the peroxisome proliferator-activated receptor gamma (PPARγ). Since the endogenous ligand of PPARγ, 15-deoxy-delta-12,14-prostaglandin J2 (15Δ-PGJ2 ), features structural similarity to DEHP and its main metabolites produced in human hepatic metabolism, mono(2-ethylhexyl) phthalate (MEHP) and mono(2-ethyl-5-oxohexyl) phthalate (MEOHP), we tested the hypothesis of direct interactions between PPARγ and DEHP or its transformation products. METHODS: Hydrogen/deuterium exchange mass spectrometry (HDX-MS) and docking were conducted to obtain structural insights into the interactions and surface plasmon resonance (SPR) analysis to reveal information about binding levels. To confirm the activation of PPARγ upon ligand binding on the cellular level, the GeneBLAzer® bioassay was performed. RESULTS: HDX-MS and SPR analyses demonstrated that the metabolites MEHP and MEOHP, but not DEHP itself, bind to the ligand binding pocket of PPARγ. This binding leads to typical activation-associated conformational changes, as observed with its endogenous ligand 15Δ-PGJ2 . Furthermore, the reporter gene assay confirmed productive interaction. DEHP was inactive up to a concentration of 14 µM, while the metabolites MEHP and MEOHP were active at low micromolar concentrations. CONCLUSIONS: In summary, this study gives structural insights into the direct interaction of PPARγ with MEHP and MEOHP and shows that the DEHP transformation products may modulate the lipid metabolism through PPARγ pathways.

Rearrangement of the Extracellular Domain/Extracellular Loop 1 Interface Is Critical for Thyrotropin Receptor Activation.

The thyroid stimulating hormone receptor (TSHR) is a G protein-coupled receptor (GPCR) with a characteristic large extracellular domain (ECD). TSHR activation is initiated by binding of the hormone ligand TSH to the ECD. How the extracellular binding event triggers the conformational changes in the transmembrane domain (TMD) necessary for intracellular G protein activation is poorly understood. To gain insight in this process, the knowledge on the relative positioning of ECD and TMD and the conformation of the linker region at the interface of ECD and TMD are of particular importance. To generate a structural model for the TSHR we applied an integrated structural biology approach combining computational techniques with experimental data. Chemical cross-linking followed by mass spectrometry yielded 17 unique distance restraints within the ECD of the TSHR, its ligand TSH, and the hormone-receptor complex. These structural restraints generally confirm the expected binding mode of TSH to the ECD as well as the general fold of the domains and were used to guide homology modeling of the ECD. Functional characterization of TSHR mutants confirms the previously suggested close proximity of Ser-281 and Ile-486 within the TSHR. Rigidifying this contact permanently with a disulfide bridge disrupts ligand-induced receptor activation and indicates that rearrangement of the ECD/extracellular loop 1 (ECL1) interface is a critical step in receptor activation. The experimentally verified contact of Ser-281 (ECD) and Ile-486 (TMD) was subsequently utilized in docking homology models of the ECD and the TMD to create a full-length model of a glycoprotein hormone receptor.

Protocols for Molecular Modeling with Rosetta3 and RosettaScripts.

Previously, we published an article providing an overview of the Rosetta suite of biomacromolecular modeling software and a series of step-by-step tutorials [Kaufmann, K. W., et al. (2010) Biochemistry 49, 2987-2998]. The overwhelming positive response to this publication we received motivates us to here share the next iteration of these tutorials that feature de novo folding, comparative modeling, loop construction, protein docking, small molecule docking, and protein design. This updated and expanded set of tutorials is needed, as since 2010 Rosetta has been fully redesigned into an object-oriented protein modeling program Rosetta3. Notable improvements include a substantially improved energy function, an XML-like language termed "RosettaScripts" for flexibly specifying modeling task, new analysis tools, the addition of the TopologyBroker to control conformational sampling, and support for multiple templates in comparative modeling. Rosetta's ability to model systems with symmetric proteins, membrane proteins, noncanonical amino acids, and RNA has also been greatly expanded and improved.

REvoLd: Ultra-Large Library Screening with an Evolutionary Algorithm in Rosetta.

Ultra-large make-on-demand compound libraries now contain billions of readily available compounds. This represents a golden opportunity for in-silico drug discovery. One challenge, however, is the time and computational cost of an exhaustive screen of such large libraries when receptor flexibility is taken into account. We propose an evolutionary algorithm to search combinatorial make-on-demand chemical space efficiently without enumerating all molecules. We exploit the feature of make-on-demand compound libraries, namely that they are constructed from lists of substrates and chemical reactions. Our novel algorithm RosettaEvolutionaryLigand (REvoLd) explores the vast search space of combinatorial libraries for protein-ligand docking with full ligand and receptor flexibility through RosettaLigand. A benchmark of REvoLd on five drug targets showed improvements in hit rates by factors between 869 and 1,622 compared to random selections. REvoLd is available as an application within the Rosetta software suite.

Community-wide evaluation of methods for predicting the effect of mutations on protein-protein interactions.

Community-wide blind prediction experiments such as CAPRI and CASP provide an objective measure of the current state of predictive methodology. Here we describe a community-wide assessment of methods to predict the effects of mutations on protein-protein interactions. Twenty-two groups predicted the effects of comprehensive saturation mutagenesis for two designed influenza hemagglutinin binders and the results were compared with experimental yeast display enrichment data obtained using deep sequencing. The most successful methods explicitly considered the effects of mutation on monomer stability in addition to binding affinity, carried out explicit side-chain sampling and backbone relaxation, evaluated packing, electrostatic, and solvation effects, and correctly identified around a third of the beneficial mutations. Much room for improvement remains for even the best techniques, and large-scale fitness landscapes should continue to provide an excellent test bed for continued evaluation of both existing and new prediction methodologies.

Computational enzyme design.

Recent developments in computational chemistry and biology have come together in the "inside-out" approach to enzyme engineering. Proteins have been designed to catalyze reactions not previously accelerated in nature. Some of these proteins fold and act as catalysts, but the success rate is still low. The achievements and limitations of the current technology are highlighted and contrasted to other protein engineering techniques. On its own, computational "inside-out" design can lead to the production of catalytically active and selective proteins, but their kinetic performances fall short of natural enzymes. When combined with directed evolution, molecular dynamics simulations, and crowd-sourced structure-prediction approaches, however, computational designs can be significantly improved in terms of binding, turnover, and thermal stability.

Integrating Expert Knowledge with Deep Learning Improves QSAR Models for CADD Modeling.

In recent years several applications of graph neural networks (GNNs) to molecular tasks have emerged. Whether GNNs outperform the traditional descriptor-based methods in the quantitative structure activity relationship (QSAR) modeling in early computer-aided drug discovery (CADD) remains an open question. This paper introduces a simple yet effective strategy to boost the predictive power of QSAR deep learning models. The strategy proposes to train GNNs together with traditional descriptors, combining the strengths of both methods. The enhanced model consistently outperforms vanilla descriptors or GNN methods on nine well-curated high throughput screening datasets over diverse therapeutic targets.

Expanding the nucleotide and sugar 1-phosphate promiscuity of nucleotidyltransferase RmlA via directed evolution.

Directed evolution is a valuable technique to improve enzyme activity in the absence of a priori structural knowledge, which can be typically enhanced via structure-guided strategies. In this study, a combination of both whole-gene error-prone polymerase chain reaction and site-saturation mutagenesis enabled the rapid identification of mutations that improved RmlA activity toward non-native substrates. These mutations have been shown to improve activities over 10-fold for several targeted substrates, including non-native pyrimidine- and purine-based NTPs as well as non-native D- and L-sugars (both α- and ß-isomers). This study highlights the first broadly applicable high throughput sugar-1-phosphate nucleotidyltransferase screen and the first proof of concept for the directed evolution of this enzyme class toward the identification of uniquely permissive RmlA variants.

Rosetta FlexPepDock to predict peptide-MHC binding: An approach for non-canonical amino acids.

Computation methods that predict the binding of peptides to MHC-I are important tools for screening and identifying immunogenic antigens and have the potential to accelerate vaccine and drug development. However, most available tools are sequence-based and optimized only for peptides containing the twenty canonical amino acids. This omits a large number of peptides containing non-canonical amino acids (NCAA), or residues that undergo varied post-translational modifications such as glycosylation or phosphorylation. These modifications fundamentally alter peptide immunogenicity. Similarly, existing structure-based methods are biased towards canonical peptide backbone structures, which may or may not be preserved when NCAAs are present. Rosetta FlexPepDock ab-initio is a structure-based computational protocol able to evaluate peptide-receptor interaction where no prior information of the peptide backbone is known. We benchmarked FlexPepDock ab-initio for docking canonical peptides to MHC-I, and illustrate for the first time the method's ability to accurately model MHC-I bound epitopes containing NCAAs. FlexPepDock ab-initio protocol was able to recapitulate near-native structures (≤1.5Å) in the top lowest-energy models for 20 out of 25 cases in our initial benchmark. Using known experimental binding affinities of twenty peptides derived from an influenza-derived peptide, we showed that FlexPepDock protocol is able to predict relative binding affinity as Rosetta energies correlate well with experimental values (r = 0.59, p = 0.006). ROC analysis revealed 80% true positive and a 40% false positive rate, with a prediction power of 93%. Finally, we demonstrate the protocol's ability to accurately recapitulate HLA-A*02:01 bound phosphopeptide backbone structures and relative binding affinity changes, the theoretical structure of the lymphocytic choriomeningitis derived glycosylated peptide GP392 bound to MHC-I H-2Db, and isolevuglandin-adducted peptides. The ability to use non-canonical amino acids in the Rosetta FlexPepDock protocol may provide useful insight into critical amino acid positions where the post-translational modification modulates immunologic responses.

Integrative model of the FSH receptor reveals the structural role of the flexible hinge region.

The follicle-stimulating hormone receptor (FSHR) belongs to the glycoprotein hormone receptors, a subfamily of G-protein-coupled receptors (GPCRs). FSHR is involved in reproductive processes such as gonadal development and maturation. Structurally, the extensive extracellular domain, which contains the hormone-binding site and is linked to the transmembrane domain by the hinge region (HR), is characteristic for these receptors. How this HR is involved in hormone binding and signal transduction is still an open question. We combined in vitro and in situ chemical crosslinking, disulfide pattern analysis, and mutation data with molecular modeling to generate experimentally driven full-length models. These models provide insights into the interface, important side-chain interactions, and activation mechanism. The interface indicates a strong involvement of the connecting loop. A major rearrangement of the HR seems implausible due to the tight arrangement and fixation by disulfide bonds. The models are expected to allow for testable hypotheses about signal transduction and drug development for GPHRs.

Efficient Sampling of Protein Loop Regions Using Conformational Hashing Complemented with Random Coordinate Descent.

De novo construction of loop regions is an important problem in computational structural biology. Compared to regions with well-defined secondary structure, loops tend to exhibit significant conformational heterogeneity. As a result, their structures are often ambiguous when determined using experimental data obtained by crystallography, cryo-EM, or NMR. Although structurally diverse models could provide a more relevant representation of proteins in their native states, obtaining large numbers of biophysically realistic and physiologically relevant loop conformations is a resource-consuming task. To address this need, we developed a novel loop construction algorithm, Hash/RCD, that combines knowledge-based conformational hashing with random coordinate descent (RCD). This hybrid approach achieved a closure rate of 100% on a benchmark set of 195 loops in 29 proteins that range from 3 to 31 residues. More importantly, the use of templates allows Hash/RCD to maintain the accuracy of state-of-the-art coordinate descent methods while reducing sampling time from over 400 to 141 ms. These results highlight how the integration of coordinate descent with knowledge-based sampling overcomes barriers inherent to either approach in isolation. This method may facilitate the identification of native-like loop conformations using experimental data or full-atom scoring functions by allowing rapid sampling of large numbers of loops. In this manuscript, we investigate and discuss the advantages, bottlenecks, and limitations of combining conformational hashing with RCD. By providing a detailed technical description of the Hash/RCD algorithm, we hope to facilitate its implementation by other researchers.

Ensuring scientific reproducibility in bio-macromolecular modeling via extensive, automated benchmarks.

Each year vast international resources are wasted on irreproducible research. The scientific community has been slow to adopt standard software engineering practices, despite the increases in high-dimensional data, complexities of workflows, and computational environments. Here we show how scientific software applications can be created in a reproducible manner when simple design goals for reproducibility are met. We describe the implementation of a test server framework and 40 scientific benchmarks, covering numerous applications in Rosetta bio-macromolecular modeling. High performance computing cluster integration allows these benchmarks to run continuously and automatically. Detailed protocol captures are useful for developers and users of Rosetta and other macromolecular modeling tools. The framework and design concepts presented here are valuable for developers and users of any type of scientific software and for the scientific community to create reproducible methods. Specific examples highlight the utility of this framework, and the comprehensive documentation illustrates the ease of adding new tests in a matter of hours.

Delineating the Molecular Basis of the Calmodulin‒bMunc13-2 Interaction by Cross-Linking/Mass Spectrometry-Evidence for a Novel CaM Binding Motif in bMunc13-2.

Exploring the interactions between the Ca2+ binding protein calmodulin (CaM) and its target proteins remains a challenging task. Members of the Munc13 protein family play an essential role in short-term synaptic plasticity, modulated via the interaction with CaM at the presynaptic compartment. In this study, we focus on the bMunc13-2 isoform expressed in the brain, as strong changes in synaptic transmission were observed upon its mutagenesis or deletion. The CaM‒bMunc13-2 interaction was previously characterized at the molecular level using short bMunc13-2-derived peptides only, revealing a classical 1‒5‒10 CaM binding motif. Using larger protein constructs, we have now identified for the first time a novel and unique CaM binding site in bMunc13-2 that contains an N-terminal extension of a classical 1‒5‒10 CaM binding motif. We characterize this motif using a range of biochemical and biophysical methods and highlight its importance for the CaM‒bMunc13-2 interaction.