Protein Design Using Physics Informed Neural Networks.

The inverse protein folding problem, also known as protein sequence design, seeks to predict an amino acid sequence that folds into a specific structure and performs a specific function. Recent advancements in machine learning techniques have been successful in generating functional sequences, outperforming previous energy function-based methods. However, these machine learning methods are limited in their interoperability and robustness, especially when designing proteins that must function under non-ambient conditions, such as high temperature, extreme pH, or in various ionic solvents. To address this issue, we propose a new Physics-Informed Neural Networks (PINNs)-based protein sequence design approach. Our approach combines all-atom molecular dynamics simulations, a PINNs MD surrogate model, and a relaxation of binary programming to solve the protein design task while optimizing both energy and the structural stability of proteins. We demonstrate the effectiveness of our design framework in designing proteins that can function under non-ambient conditions.

The Peptidyl-Prolyl cis-trans isomerase, Pin1, associates with Protein Kinase C θ via a critical Phospho-Thr-Pro motif in the V3 regulatory domain.

Protein kinase C-θ (PKCθ) is a member of the novel PKC subfamily known for its selective and predominant expression in T lymphocytes where it regulates essential functions required for T cell activation and proliferation. Our previous studies provided a mechanistic explanation for the recruitment of PKCθ to the center of the immunological synapse (IS) by demonstrating that a proline-rich (PR) motif within the V3 region in the regulatory domain of PKCθ is necessary and sufficient for PKCθ IS localization and function. Herein, we highlight the importance of Thr335-Pro residue in the PR motif, the phosphorylation of which is key in the activation of PKCθ and its subsequent IS localization. We demonstrate that the phospho-Thr335-Pro motif serves as a putative binding site for the peptidyl-prolyl cis-trans isomerase (PPIase), Pin1, an enzyme that specifically recognizes peptide bonds at phospho-Ser/Thr-Pro motifs. Binding assays revealed that mutagenesis of PKCθ-Thr335-to-Ala abolished the ability of PKCθ to interact with Pin1, while Thr335 replacement by a Glu phosphomimetic, restored PKCθ binding to Pin1, suggesting that Pin1-PKCθ association is contingent upon the phosphorylation of the PKCθ-Thr335-Pro motif. Similarly, the Pin1 mutant, R17A, failed to associate with PKCθ, suggesting that the integrity of the Pin1 N-terminal WW domain is a requisite for Pin1-PKCθ interaction. In silico docking studies underpinned the role of critical residues in the Pin1-WW domain and the PKCθ phospho-Thr335-Pro motif, to form a stable interaction between Pin1 and PKCθ. Furthermore, TCR crosslinking in human Jurkat T cells and C57BL/6J mouse-derived splenic T cells promoted a rapid and transient formation of Pin1-PKCθ complexes, which followed a T cell activation-dependent temporal kinetic, suggesting a role for Pin1 in PKCθ-dependent early activation events in TCR-triggered T cells. PPIases that belong to other subfamilies, i.e., cyclophilin A or FK506-binding protein, failed to associate with PKCθ, indicating the specificity of the Pin1-PKCθ association. Fluorescent cell staining and imaging analyses demonstrated that TCR/CD3 triggering promotes the colocalization of PKCθ and Pin1 at the cell membrane. Furthermore, interaction of influenza hemagglutinin peptide (HA307-319)-specific T cells with antigen-fed antigen presenting cells (APCs) led to colocalization of PKCθ and Pin1 at the center of the IS. Together, we point to an uncovered function for the Thr335-Pro motif within the PKCθ-V3 regulatory domain to serve as a priming site for its activation upon phosphorylation and highlight its tenability to serve as a regulatory site for the Pin1 cis-trans isomerase.

Mimetic Neural Networks: A Unified Framework for Protein Design and Folding.

Recent advancements in machine learning techniques for protein structure prediction motivate better results in its inverse problem-protein design. In this work we introduce a new graph mimetic neural network, MimNet, and show that it is possible to build a reversible architecture that solves the structure and design problems in tandem, allowing to improve protein backbone design when the structure is better estimated. We use the ProteinNet data set and show that the state of the art results in protein design can be met and even improved, given recent architectures for protein folding.

Estimation of model accuracy by a unique set of features and tree-based regressor.

Computationally generated models of protein structures bridge the gap between the practically negligible price tag of sequencing and the high cost of experimental structure determination. By providing a low-cost (and often free) partial alternative to experimentally determined structures, these models help biologists design and interpret their experiments. Obviously, the more accurate the models the more useful they are. However, methods for protein structure prediction generate many structural models of various qualities, necessitating means for the estimation of their accuracy. In this work we present MESHI_consensus, a new method for the estimation of model accuracy. The method uses a tree-based regressor and a set of structural, target-based, and consensus-based features. The new method achieved high performance in the EMA (Estimation of Model Accuracy) track of the recent CASP14 community-wide experiment ( https://predictioncenter.org/casp14/index.cgi ). The tertiary structure prediction track of that experiment revealed an unprecedented leap in prediction performance by a single prediction group/method, namely AlphaFold2. This achievement would inevitably have a profound impact on the field of protein structure prediction, including the accuracy estimation sub-task. We conclude this manuscript with some speculations regarding the future role of accuracy estimation in a new era of accurate protein structure prediction.

Predicting mutant outcome by combining deep mutational scanning and machine learning.

Deep mutational scanning provides unprecedented wealth of quantitative data regarding the functional outcome of mutations in proteins. A single experiment may measure properties (eg, structural stability) of numerous protein variants. Leveraging the experimental data to gain insights about unexplored regions of the mutational landscape is a major computational challenge. Such insights may facilitate further experimental work and accelerate the development of novel protein variants with beneficial therapeutic or industrially relevant properties. Here we present a novel, machine learning approach for the prediction of functional mutation outcome in the context of deep mutational screens. Using sequence (one-hot) features of variants with known properties, as well as structural features derived from models thereof, we train predictive statistical models to estimate the unknown properties of other variants. The utility of the new computational scheme is demonstrated using five sets of mutational scanning data, denoted "targets": (a) protease specificity of APPI (amyloid precursor protein inhibitor) variants; (b-d) three stability related properties of IGBPG (immunoglobulin G-binding ß1 domain of streptococcal protein G) variants; and (e) fluorescence of GFP (green fluorescent protein) variants. Performance is measured by the overall correlation of the predicted and observed properties, and enrichment-the ability to predict the most potent variants and presumably guide further experiments. Despite the diversity of the targets the statistical models can generalize variant examples thereof and predict the properties of test variants with both single and multiple mutations.

Unraveling the hidden role of a uORF-encoded peptide as a kinase inhibitor of PKCs.

Approximately 40% of human messenger RNAs (mRNAs) contain upstream open reading frames (uORFs) in their 5' untranslated regions. Some of these uORF sequences, thought to attenuate scanning ribosomes or lead to mRNA degradation, were recently shown to be translated, although the function of the encoded peptides remains unknown. Here, we show a uORF-encoded peptide that exhibits kinase inhibitory functions. This uORF, upstream of the protein kinase C-eta (PKC-η) main ORF, encodes a peptide (uPEP2) containing the typical PKC pseudosubstrate motif present in all PKCs that autoinhibits their kinase activity. We show that uPEP2 directly binds to and selectively inhibits the catalytic activity of novel PKCs but not of classical or atypical PKCs. The endogenous deletion of uORF2 or its overexpression in MCF-7 cells revealed that the endogenously translated uPEP2 reduces the protein levels of PKC-η and other novel PKCs and restricts cell proliferation. Functionally, treatment of breast cancer cells with uPEP2 diminished cell survival and their migration and synergized with chemotherapy by interfering with the response to DNA damage. Furthermore, in a xenograft of MDA-MB-231 breast cancer tumor in mice models, uPEP2 suppressed tumor progression, invasion, and metastasis. Tumor histology showed reduced proliferation, enhanced cell death, and lower protein expression levels of novel PKCs along with diminished phosphorylation of PKC substrates. Hence, our study demonstrates that uORFs may encode biologically active peptides beyond their role as translation regulators of their downstream ORFs. Together, we point to a unique function of a uORF-encoded peptide as a kinase inhibitor, pertinent to cancer therapy.

Redundancy-weighting the PDB for detailed secondary structure prediction using deep-learning models.

MOTIVATION: The Protein Data Bank (PDB), the ultimate source for data in structural biology, is inherently imbalanced. To alleviate biases, virtually all structural biology studies use nonredundant (NR) subsets of the PDB, which include only a fraction of the available data. An alternative approach, dubbed redundancy-weighting (RW), down-weights redundant entries rather than discarding them. This approach may be particularly helpful for machine-learning (ML) methods that use the PDB as their source for data. Methods for secondary structure prediction (SSP) have greatly improved over the years with recent studies achieving above 70% accuracy for eight-class (DSSP) prediction. As these methods typically incorporate ML techniques, training on RW datasets might improve accuracy, as well as pave the way toward larger and more informative secondary structure classes. RESULTS: This study compares the SSP performances of deep-learning models trained on either RW or NR datasets. We show that training on RW sets consistently results in better prediction of 3- (HCE), 8- (DSSP) and 13-class (STR2) secondary structures. AVAILABILITY AND IMPLEMENTATION: The ML models, the datasets used for their derivation and testing, and a stand-alone SSP program for DSSP and STR2 predictions, are freely available under LGPL license in http://meshi1.cs.bgu.ac.il/rw. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Purely Structural Protein Scoring Functions Using Support Vector Machine and Ensemble Learning.

The function of a protein is determined by its structure, which creates a need for efficient methods of protein structure determination to advance scientific and medical research. Because current experimental structure determination methods carry a high price tag, computational predictions are highly desirable. Given a protein sequence, computational methods produce numerous 3D structures known as decoys. Selection of the best quality decoys is both challenging and essential as the end users can handle only a few ones. Therefore, scoring functions are central to decoy selection. They combine measurable features into a single number indicator of decoy quality. Unfortunately, current scoring functions do not consistently select the best decoys. Machine learning techniques offer great potential to improve decoy scoring. This paper presents two machine-learning based scoring functions to predict the quality of proteins structures, i.e., the similarity between the predicted structure and the experimental one without knowing the latter. We use different metrics to compare these scoring functions against three state-of-the-art scores. This is a first attempt at comparing different scoring functions using the same non-redundant dataset for training and testing and the same features. The results show that adding informative features may be more significant than the method used.

An analysis and evaluation of the WeFold collaborative for protein structure prediction and its pipelines in CASP11 and CASP12.

Every two years groups worldwide participate in the Critical Assessment of Protein Structure Prediction (CASP) experiment to blindly test the strengths and weaknesses of their computational methods. CASP has significantly advanced the field but many hurdles still remain, which may require new ideas and collaborations. In 2012 a web-based effort called WeFold, was initiated to promote collaboration within the CASP community and attract researchers from other fields to contribute new ideas to CASP. Members of the WeFold coopetition (cooperation and competition) participated in CASP as individual teams, but also shared components of their methods to create hybrid pipelines and actively contributed to this effort. We assert that the scale and diversity of integrative prediction pipelines could not have been achieved by any individual lab or even by any collaboration among a few partners. The models contributed by the participating groups and generated by the pipelines are publicly available at the WeFold website providing a wealth of data that remains to be tapped. Here, we analyze the results of the 2014 and 2016 pipelines showing improvements according to the CASP assessment as well as areas that require further adjustments and research.

Methods for estimation of model accuracy in CASP12.

Methods to reliably estimate the quality of 3D models of proteins are essential drivers for the wide adoption and serious acceptance of protein structure predictions by life scientists. In this article, the most successful groups in CASP12 describe their latest methods for estimates of model accuracy (EMA). We show that pure single model accuracy estimation methods have shown clear progress since CASP11; the 3 top methods (MESHI, ProQ3, SVMQA) all perform better than the top method of CASP11 (ProQ2). Although the pure single model accuracy estimation methods outperform quasi-single (ModFOLD6 variations) and consensus methods (Pcons, ModFOLDclust2, Pcomb-domain, and Wallner) in model selection, they are still not as good as those methods in absolute model quality estimation and predictions of local quality. Finally, we show that when using contact-based model quality measures (CAD, lDDT) the single model quality methods perform relatively better.

Redundancy-weighting for better inference of protein structural features.

MOTIVATION: Structural knowledge, extracted from the Protein Data Bank (PDB), underlies numerous potential functions and prediction methods. The PDB, however, is highly biased: many proteins have more than one entry, while entire protein families are represented by a single structure, or even not at all. The standard solution to this problem is to limit the studies to non-redundant subsets of the PDB. While alleviating biases, this solution hides the many-to-many relations between sequences and structures. That is, non-redundant datasets conceal the diversity of sequences that share the same fold and the existence of multiple conformations for the same protein. A particularly disturbing aspect of non-redundant subsets is that they hardly benefit from the rapid pace of protein structure determination, as most newly solved structures fall within existing families. RESULTS: In this study we explore the concept of redundancy-weighted datasets, originally suggested by Miyazawa and Jernigan. Redundancy-weighted datasets include all available structures and associate them (or features thereof) with weights that are inversely proportional to the number of their homologs. Here, we provide the first systematic comparison of redundancy-weighted datasets with non-redundant ones. We test three weighting schemes and show that the distributions of structural features that they produce are smoother (having higher entropy) compared with the distributions inferred from non-redundant datasets. We further show that these smoothed distributions are both more robust and more correct than their non-redundant counterparts. We suggest that the better distributions, inferred using redundancy-weighting, may improve the accuracy of knowledge-based potentials and increase the power of protein structure prediction methods. Consequently, they may enhance model-driven molecular biology.

WeFold: a coopetition for protein structure prediction.

The protein structure prediction problem continues to elude scientists. Despite the introduction of many methods, only modest gains were made over the last decade for certain classes of prediction targets. To address this challenge, a social-media based worldwide collaborative effort, named WeFold, was undertaken by 13 labs. During the collaboration, the laboratories were simultaneously competing with each other. Here, we present the first attempt at "coopetition" in scientific research applied to the protein structure prediction and refinement problems. The coopetition was possible by allowing the participating labs to contribute different components of their protein structure prediction pipelines and create new hybrid pipelines that they tested during CASP10. This manuscript describes both successes and areas needing improvement as identified throughout the first WeFold experiment and discusses the efforts that are underway to advance this initiative. A footprint of all contributions and structures are publicly accessible at http://www.wefold.org.

Identification of a colonial chordate histocompatibility gene.

Histocompatibility is the basis by which multicellular organisms of the same species distinguish self from nonself. Relatively little is known about the mechanisms underlying histocompatibility reactions in lower organisms. Botryllus schlosseri is a colonial urochordate, a sister group of vertebrates, that exhibits a genetically determined natural transplantation reaction, whereby self-recognition between colonies leads to formation of parabionts with a common vasculature, whereas rejection occurs between incompatible colonies. Using genetically defined lines, whole-transcriptome sequencing, and genomics, we identified a single gene that encodes self-nonself and determines "graft" outcomes in this organism. This gene is significantly up-regulated in colonies poised to undergo fusion and/or rejection, is highly expressed in the vasculature, and is functionally linked to histocompatibility outcomes. These findings establish a platform for advancing the science of allorecognition.

Knowledge-based potential for positioning membrane-associated structures and assessing residue-specific energetic contributions.

The complex hydrophobic and hydrophilic milieus of membrane-associated proteins pose experimental and theoretical challenges to their understanding. Here, we produce a nonredundant database to compute knowledge-based asymmetric cross-membrane potentials from the per-residue distributions of C(ß), C(γ) and functional group atoms. We predict transmembrane and peripherally associated regions from genomic sequence and position peptides and protein structures relative to the bilayer (available at http://www.degradolab.org/ez). The pseudo-energy topological landscapes underscore positional stability and functional mechanisms demonstrated here for antimicrobial peptides, transmembrane proteins, and viral fusion proteins. Moreover, experimental effects of point mutations on the relative ratio changes of dual-topology proteins are quantitatively reproduced. The functional group potential and the membrane-exposed residues display the largest energetic changes enabling to detect native-like structures from decoys. Hence, focusing on the uniqueness of membrane-associated proteins and peptides, we quantitatively parameterize their cross-membrane propensity, thus facilitating structural refinement, characterization, prediction, and design.

Hierarchical self-organization of cytoskeletal active networks.

The structural reorganization of the actin cytoskeleton is facilitated through the action of motor proteins that crosslink the actin filaments and transport them relative to each other. Here, we present a combined experimental-computational study that probes the dynamic evolution of mixtures of actin filaments and clusters of myosin motors. While on small spatial and temporal scales the system behaves in a very noisy manner, on larger scales it evolves into several well distinct patterns such as bundles, asters and networks. These patterns are characterized by junctions with high connectivity, whose formation is possible due to the organization of the motors in 'oligoclusters' (intermediate-size aggregates). The simulations reveal that the self-organization process proceeds through a series of hierarchical steps, starting from local microscopic moves and ranging up to the macroscopic large scales where the steady-state structures are formed. Our results shed light on the mechanisms involved in processes such as cytokinesis and cellular contractility, where myosin motors organized in clusters operate cooperatively to induce the structural organization of cytoskeletal networks.

A library of protein surface patches discriminates between native structures and decoys generated by structure prediction servers.

BACKGROUND: Protein surfaces serve as an interface with the molecular environment and are thus tightly bound to protein function. On the surface, geometric and chemical complementarity to other molecules provides interaction specificity for ligand binding, docking of bio-macromolecules, and enzymatic catalysis.As of today, there is no accepted general scheme to represent protein surfaces. Furthermore, most of the research on protein surface focuses on regions of specific interest such as interaction, ligand binding, and docking sites. We present a first step toward a general purpose representation of protein surfaces: a novel surface patch library that represents most surface patches (~98%) in a data set regardless of their functional roles. RESULTS: Surface patches, in this work, are small fractions of the protein surface. Using a measure of inter-patch distance, we clustered patches extracted from a data set of high quality, non-redundant, proteins. The surface patch library is the collection of all the cluster centroids; thus, each of the data set patches is close to one of the elements in the library.We demonstrate the biological significance of our method through the ability of the library to capture surface characteristics of native protein structures as opposed to those of decoy sets generated by state-of-the-art protein structure prediction methods. The patches of the decoys are significantly less compatible with the library than their corresponding native structures, allowing us to reliably distinguish native models from models generated by servers. This trend, however, does not extend to the decoys themselves, as their similarity to the native structures does not correlate with compatibility with the library. CONCLUSIONS: We expect that this high-quality, generic surface patch library will add a new perspective to the description of protein structures and improve our ability to predict them. In particular, we expect that it will help improve the prediction of surface features that are apparently neglected by current techniques.The surface patch libraries are publicly available at http://www.cs.bgu.ac.il/~keasar/patchLibrary.

Structure-based identification of catalytic residues.

The identification of catalytic residues is an essential step in functional characterization of enzymes. We present a purely structural approach to this problem, which is motivated by the difficulty of evolution-based methods to annotate structural genomics targets that have few or no homologs in the databases. Our approach combines a state-of-the-art support vector machine (SVM) classifier with novel structural features that augment structural clues by spatial averaging and Z scoring. Special attention is paid to the class imbalance problem that stems from the overwhelming number of non-catalytic residues in enzymes compared to catalytic residues. This problem is tackled by: (1) optimizing the classifier to maximize a performance criterion that considers both Type I and Type II errors in the classification of catalytic and non-catalytic residues; (2) under-sampling non-catalytic residues before SVM training; and (3) during SVM training, penalizing errors in learning catalytic residues more than errors in learning non-catalytic residues. Tested on four enzyme datasets, one specifically designed by us to mimic the structural genomics scenario and three previously evaluated datasets, our structure-based classifier is never inferior to similar structure-based classifiers and comparable to classifiers that use both structural and evolutionary features. In addition to the evaluation of the performance of catalytic residue identification, we also present detailed case studies on three proteins. This analysis suggests that many false positive predictions may correspond to binding sites and other functional residues. A web server that implements the method, our own-designed database, and the source code of the programs are publicly available at http://www.cs.bgu.ac.il/∼meshi/functionPrediction.

Enhancement of beta-sheet assembly by cooperative hydrogen bonds potential.

MOTIVATION: The roughness of energy landscapes is a major obstacle to protein structure prediction, since it forces conformational searches to spend much time struggling to escape numerous traps. Specifically, beta-sheet formation is prone to stray, since many possible combinations of hydrogen bonds are dead ends in terms of beta-sheet assembly. It has been shown that cooperative terms for backbone hydrogen bonds ease this problem by augmenting hydrogen bond patterns that are consistent with beta sheets. Here, we present a novel cooperative hydrogen-bond term that is both effective in promoting beta sheets and computationally efficient. In addition, the new term is differentiable and operates on all-atom protein models. RESULTS: Energy optimization of poly-alanine chains under the new term led to significantly more beta-sheet content than optimization under a non-cooperative term. Furthermore, the optimized structure included very few non-native patterns. AVAILABILITY: The new term is implemented within the MESHI package and is freely available at http://cs.bgu.ac.il/ approximately meshi.

Identification of the Zn2+ binding site and mode of operation of a mammalian Zn2+ transporter.

Vesicular zinc transporters (ZnTs) play a critical role in regulating Zn2+ homeostasis in various cellular compartments and are linked to major diseases ranging from Alzheimer disease to diabetes. Despite their importance, the intracellular localization of ZnTs poses a major challenge for establishing the mechanisms by which they function and the identity of their ion binding sites. Here, we combine fluorescence-based functional analysis and structural modeling aimed at elucidating these functional aspects. Expression of ZnT5 was followed by both accelerated removal of Zn2+ from the cytoplasm and its increased vesicular sequestration. Further, activity of this zinc transport was coupled to alkalinization of the trans-Golgi network. Finally, structural modeling of ZnT5, based on the x-ray structure of the bacterial metal transporter YiiP, identified four residues that can potentially form the zinc binding site on ZnT5. Consistent with this model, replacement of these residues, Asp599 and His451, with alanine was sufficient to block Zn2+ transport. These findings indicate, for the first time, that Zn2+ transport mediated by a mammalian ZnT is catalyzed by H+/Zn2+ exchange and identify the zinc binding site of ZnT proteins essential for zinc transport.

Prediction of structural stability of short beta-hairpin peptides by molecular dynamics and knowledge-based potentials.

BACKGROUND: The structural stability of peptides in solution strongly affects their binding affinities and specificities. Thus, in peptide biotechnology, an increase in the structural stability is often desirable. The present work combines two orthogonal computational techniques, Molecular Dynamics and a knowledge-based potential, for the prediction of structural stability of short peptides (< 20 residues) in solution. RESULTS: We tested the new approach on four families of short beta-hairpin peptides: TrpZip, MBH, bhpW and EPO, whose structural stabilities have been experimentally measured in previous studies. For all four families, both computational techniques show considerable correlation (r > 0.65) with the experimentally measured stabilities. The consensus of the two techniques shows higher correlation (r > 0.82). CONCLUSION: Our results suggest a prediction scheme that can be used to estimate the relative structural stability within a peptide family. We discuss the applicability of this predictive approach for in-silico screening of combinatorial peptide libraries.