Computational redesign of a fluorogen activating protein with Rosetta.

The use of unnatural fluorogenic molecules widely expands the pallet of available genetically encoded fluorescent imaging tools through the design of fluorogen activating proteins (FAPs). While there is already a handful of such probes available, each of them went through laborious cycles of in vitro screening and selection. Computational modeling approaches are evolving incredibly fast right now and are demonstrating great results in many applications, including de novo protein design. It suggests that the easier task of fine-tuning the fluorogen-binding properties of an already functional protein in silico should be readily achievable. To test this hypothesis, we used Rosetta for computational ligand docking followed by protein binding pocket redesign to further improve the previously described FAP DiB1 that is capable of binding to a BODIPY-like dye M739. Despite an inaccurate initial docking of the chromophore, the incorporated mutations nevertheless improved multiple photophysical parameters as well as the overall performance of the tag. The designed protein, DiB-RM, shows higher brightness, localization precision, and apparent photostability in protein-PAINT super-resolution imaging compared to its parental variant DiB1. Moreover, DiB-RM can be cleaved to obtain an efficient split system with enhanced performance compared to a parental DiB-split system. The possible reasons for the inaccurate ligand binding pose prediction and its consequence on the outcome of the design experiment are further discussed.

ProDCoNN: Protein design using a convolutional neural network.

Designing protein sequences that fold to a given three-dimensional (3D) structure has long been a challenging problem in computational structural biology with significant theoretical and practical implications. In this study, we first formulated this problem as predicting the residue type given the 3D structural environment around the C α atom of a residue, which is repeated for each residue of a protein. We designed a nine-layer 3D deep convolutional neural network (CNN) that takes as input a gridded box with the atomic coordinates and types around a residue. Several CNN layers were designed to capture structure information at different scales, such as bond lengths, bond angles, torsion angles, and secondary structures. Trained on a very large number of protein structures, the method, called ProDCoNN (protein design with CNN), achieved state-of-the-art performance when tested on large numbers of test proteins and benchmark datasets.

Optimizing antibody affinity and stability by the automated design of the variable light-heavy chain interfaces.

Antibodies developed for research and clinical applications may exhibit suboptimal stability, expressibility, or affinity. Existing optimization strategies focus on surface mutations, whereas natural affinity maturation also introduces mutations in the antibody core, simultaneously improving stability and affinity. To systematically map the mutational tolerance of an antibody variable fragment (Fv), we performed yeast display and applied deep mutational scanning to an anti-lysozyme antibody and found that many of the affinity-enhancing mutations clustered at the variable light-heavy chain interface, within the antibody core. Rosetta design combined enhancing mutations, yielding a variant with tenfold higher affinity and substantially improved stability. To make this approach broadly accessible, we developed AbLIFT, an automated web server that designs multipoint core mutations to improve contacts between specific Fv light and heavy chains (http://AbLIFT.weizmann.ac.il). We applied AbLIFT to two unrelated antibodies targeting the human antigens VEGF and QSOX1. Strikingly, the designs improved stability, affinity, and expression yields. The results provide proof-of-principle for bypassing laborious cycles of antibody engineering through automated computational affinity and stability design.

Computational design of thermostabilizing point mutations for G protein-coupled receptors.

Engineering of GPCR constructs with improved thermostability is a key for successful structural and biochemical studies of this transmembrane protein family, targeted by 40% of all therapeutic drugs. Here we introduce a comprehensive computational approach to effective prediction of stabilizing mutations in GPCRs, named CompoMug, which employs sequence-based analysis, structural information, and a derived machine learning predictor. Tested experimentally on the serotonin 5-HT2C receptor target, CompoMug predictions resulted in 10 new stabilizing mutations, with an apparent thermostability gain ~8.8°C for the best single mutation and ~13°C for a triple mutant. Binding of antagonists confers further stabilization for the triple mutant receptor, with total gains of ~21°C as compared to wild type apo 5-HT2C. The predicted mutations enabled crystallization and structure determination for the 5-HT2C receptor complexes in inactive and active-like states. While CompoMug already shows high 25% hit rate and utility in GPCR structural studies, further improvements are expected with accumulation of structural and mutation data.

RosettaAntibodyDesign (RAbD): A general framework for computational antibody design.

A structural-bioinformatics-based computational methodology and framework have been developed for the design of antibodies to targets of interest. RosettaAntibodyDesign (RAbD) samples the diverse sequence, structure, and binding space of an antibody to an antigen in highly customizable protocols for the design of antibodies in a broad range of applications. The program samples antibody sequences and structures by grafting structures from a widely accepted set of the canonical clusters of CDRs (North et al., J. Mol. Biol., 406:228-256, 2011). It then performs sequence design according to amino acid sequence profiles of each cluster, and samples CDR backbones using a flexible-backbone design protocol incorporating cluster-based CDR constraints. Starting from an existing experimental or computationally modeled antigen-antibody structure, RAbD can be used to redesign a single CDR or multiple CDRs with loops of different length, conformation, and sequence. We rigorously benchmarked RAbD on a set of 60 diverse antibody-antigen complexes, using two design strategies-optimizing total Rosetta energy and optimizing interface energy alone. We utilized two novel metrics for measuring success in computational protein design. The design risk ratio (DRR) is equal to the frequency of recovery of native CDR lengths and clusters divided by the frequency of sampling of those features during the Monte Carlo design procedure. Ratios greater than 1.0 indicate that the design process is picking out the native more frequently than expected from their sampled rate. We achieved DRRs for the non-H3 CDRs of between 2.4 and 4.0. The antigen risk ratio (ARR) is the ratio of frequencies of the native amino acid types, CDR lengths, and clusters in the output decoys for simulations performed in the presence and absence of the antigen. For CDRs, we achieved cluster ARRs as high as 2.5 for L1 and 1.5 for H2. For sequence design simulations without CDR grafting, the overall recovery for the native amino acid types for residues that contact the antigen in the native structures was 72% in simulations performed in the presence of the antigen and 48% in simulations performed without the antigen, for an ARR of 1.5. For the non-contacting residues, the ARR was 1.08. This shows that the sequence profiles are able to maintain the amino acid types of these conserved, buried sites, while recovery of the exposed, contacting residues requires the presence of the antigen-antibody interface. We tested RAbD experimentally on both a lambda and kappa antibody-antigen complex, successfully improving their affinities 10 to 50 fold by replacing individual CDRs of the native antibody with new CDR lengths and clusters.

Effective Design of Multifunctional Peptides by Combining Compatible Functions.

Multifunctionality is a common trait of many natural proteins and peptides, yet the rules to generate such multifunctionality remain unclear. We propose that the rules defining some protein/peptide functions are compatible. To explore this hypothesis, we trained a computational method to predict cell-penetrating peptides at the sequence level and learned that antimicrobial peptides and DNA-binding proteins are compatible with the rules of our predictor. Based on this finding, we expected that designing peptides for CPP activity may render AMP and DNA-binding activities. To test this prediction, we designed peptides that embedded two independent functional domains (nuclear localization and yeast pheromone activity), linked by optimizing their composition to fit the rules characterizing cell-penetrating peptides. These peptides presented effective cell penetration, DNA-binding, pheromone and antimicrobial activities, thus confirming the effectiveness of our computational approach to design multifunctional peptides with potential therapeutic uses. Our computational implementation is available at http://bis.ifc.unam.mx/en/software/dcf.

Structure-guided deimmunization of therapeutic proteins.

Therapeutic proteins continue to yield revolutionary new treatments for a growing spectrum of human disease, but the development of these powerful drugs requires solving a unique set of challenges. For instance, it is increasingly apparent that mitigating potential anti-therapeutic immune responses, driven by molecular recognition of a therapeutic protein's peptide fragments, may be best accomplished early in the drug development process. One may eliminate immunogenic peptide fragments by mutating the cognate amino acid sequences, but deimmunizing mutations are constrained by the need for a folded, stable, and functional protein structure. These two concerns may be competing, as the mutations that are best at reducing immunogenicity often involve amino acids that are substantially different physicochemically. We develop a novel approach, called EpiSweep, that simultaneously optimizes both concerns. Our algorithm identifies sets of mutations making such Pareto optimal trade-offs between structure and immunogenicity, embodied by a molecular mechanics energy function and a T-cell epitope predictor, respectively. EpiSweep integrates structure-based protein design, sequence-based protein deimmunization, and algorithms for finding the Pareto frontier of a design space. While structure-based protein design is NP-hard, we employ integer programming techniques that are efficient in practice. Furthermore, EpiSweep only invokes the optimizer once per identified Pareto optimal design. We show that EpiSweep designs of regions of the therapeutics erythropoietin and staphylokinase are predicted to outperform previous experimental efforts. We also demonstrate EpiSweep's capacity for deimmunization of the entire proteins, case analyses involving dozens of predicted epitopes, and tens of thousands of unique side-chain interactions. Ultimately, Epi-Sweep is a powerful protein design tool that guides the protein engineer toward the most promising immunotolerant biotherapeutic candidates.

MAP(2.0)3D: a sequence/structure based server for protein engineering.

The Mutagenesis Assistant Program (MAP) is a web-based tool to provide statistical analyses of the mutational biases of directed evolution experiments on amino acid substitution patterns. MAP analysis assists protein engineers in the benchmarking of random mutagenesis methods that generate single nucleotide mutations in a codon. Herein, we describe a completely renewed and improved version of the MAP server, the MAP(2.0)3D server, which correlates the generated amino acid substitution patterns to the structural information of the target protein. This correlation aids in the selection of a more suitable random mutagenesis method with specific biases on amino acid substitution patterns. In particular, the new server represents MAP indicators on secondary and tertiary structure and correlates them to specific structural components such as hydrogen bonds, hydrophobic contacts, salt bridges, solvent accessibility, and crystallographic B-factors. Three model proteins (D-amino oxidase, phytase, and N-acetylneuraminic acid aldolase) are used to illustrate the novel capability of the server. MAP(2.0)3D server is available publicly at http://map.jacobs-university.de/map3d.html.

Optimization of therapeutic proteins to delete T-cell epitopes while maintaining beneficial residue interactions.

Exogenous enzymes, signaling peptides, and other classes of nonhuman proteins represent a potentially massive but largely untapped pool of biotherapeutic agents. Adapting a foreign protein for therapeutic use poses numerous design challenges. We focus here on one significant problem: modifying the protein to mitigate the immune response mounted against "non-self" proteins, while not adversely affecting the protein's stability or therapeutic activity. In order to propose such variants suitable for experimental evaluation, this paper develops a computational method to select sets of mutations predicted to delete immunogenic T-cell epitopes, as evaluated by a 9-mer potential, while simultaneously maintaining important residues and residue interactions, as evaluated by one- and two-body potentials. While this design problem is NP-hard, we develop an integer programming approach that works very well in practice. We demonstrate the effectiveness of our approach by developing plans for biotherapeutic proteins that, in previous studies, have been partially deimmunized via extensive experimental characterization and modification of limited segments. In contrast, our global optimization technique considers an entire protein and accounts for all residues, residue interactions, and epitopes in proposing candidates worth subjecting to experimental evaluation.

Reverse engineering the evolution of protein interaction networks.

Protein interaction network analyses have moved beyond simple topological observations to functional and evolutionary inferences based on the construction of putative ancestral networks. Evolutionary studies of protein interaction networks are generally derived from network comparisons, are limited in scope, or are theoretic dynamic models that aren't contextualized to an organism's specific genes. A biologically faithful network evolution reconstruction which ties evolution of the network itself to the actual genes of an organism would help fill in the evolutionary gaps between the gene network "snapshots" of evolution we have from different species today. Here we present a novel framework for reverse engineering the evolution of protein interaction networks of extant species using phylogenetic gene trees and protein interaction data. We applied the framework to Saccharomyces cerevisiae data and present topological trends in the evolutionary lineage of yeast.

Building a BioChemformatics database.

The structural registration of chemically modified macromolecules is vital for the development of biopharmaceuticals. However, registration and search of such complex molecules has so far posed formidable challenges performance-wise, since today's chemistry-oriented databases do not scale well to macromolecules. As a practical consequence, macromolecules tend to be stored in protein databases with a focus on protein sequence only, and salient chemistry details are therefore lost. This article describes protein format extensions and the use of pseudoatoms for representing natural amino acids in chemical structures to allow high-performance registration and retrieval of large macromolecules. The representations include exact chemical modifications and enable lossless conversion between chemistry and sequence formats. Registration is done in parallel in both sequence and chemistry formats, and users can register and retrieve molecules in either format as they choose, resulting in what we call a BioChemformatics database. Having both sequence and chemistry formats available on-demand allows for the construction of protein SAR tables with mixed sequence and chemistry information. Likewise, searching may combine sequence and chemistry terms and be performed in standard vendor applications like MDL's ISIS/Base or in-house applications using standard SQL queries.

Bioinformatics-driven, rational engineering of protein thermostability.

A longstanding goal in protein engineering is to identify specific sequence changes that endow proteins with desired functional properties. As opposed to traditional rational and random protein engineering techniques, we have employed a bioinformatic approach to identify specific sequence changes that influence key functional properties of a protein within a defined superfamily. Specifically, we have used the Bayesian sequence-based algorithms PROBE and Classifier to identify a strand-turn-strand motif that contributes to thermophilicity among members of the serine protease subtilase superfamily. By replacing a 16 amino acid sequence in the mesophilic subtilisin E (from Bacillus subtilis) with a bioinformatics-generated thermophilic model sequence, the melting temperature of subtilisin E was increased by 13 degrees C. While wild-type subtilisin E was inactive at 90 degrees C, the mutant retained a substantial fraction of its function, with ca. one-third of the activity that it has at 45 degrees C.

Reverse engineering genetic networks using evolutionary computation.

This paper proposes an improved evolutionary method for constructing the underlying network structure and inferring effective kinetic parameters from the time series data of gene expression using decoupled S-system formalism. We employed Trigonometric Differential Evolution (TDE) as the optimization engine of our algorithm for capturing the dynamics in gene expression data. A more effective fitness function for attaining the sparse structure, which is the hallmark of biological networks, has been applied. Experiments on artificial genetic network show the power of the algorithm in constructing the network structure and predicting the regulatory parameters. The method is used to evaluate interactions between genes in the SOS signaling pathway in Escherichia coli using gene expression data.

Comparative binding energy analysis of the substrate specificity of haloalkane dehalogenase from Xanthobacter autotrophicus GJ10.

Comparative binding energy (COMBINE) analysis was conducted for 18 substrates of the haloalkane dehalogenase from Xanthobacter autotrophicus GJ10 (DhlA): 1-chlorobutane, 1-chlorohexane, dichloromethane, 1,2-dichloroethane, 1,2-dichloropropane, 2-chloroethanol, epichlorohydrine, 2-chloroacetonitrile, 2-chloroacetamide, and their brominated analogues. The purpose of the COMBINE analysis was to identify the amino acid residues determining the substrate specificity of the haloalkane dehalogenase. This knowledge is essential for the tailoring of this enzyme for biotechnological applications. Complexes of the enzyme with these substrates were modeled and then refined by molecular mechanics energy minimization. The intermolecular enzyme-substrate energy was decomposed into residue-wise van der Waals and electrostatic contributions and complemented by surface area dependent and electrostatic desolvation terms. Partial least-squares projection to latent structures analysis was then used to establish relationships between the energy contributions and the experimental apparent dissociation constants. A model containing van der Waals and electrostatic intermolecular interaction energy contributions calculated using the AMBER force field explained 91% (73% cross-validated) of the quantitative variance in the apparent dissociation constants. A model based on van der Waals intermolecular contributions from AMBER and electrostatic interactions derived from the Poisson-Boltzmann equation explained 93% (74% cross-validated) of the quantitative variance. COMBINE models predicted correctly the change in apparent dissociation constants upon single-point mutation of DhlA for six enzyme-substrate complexes. The amino acid residues contributing most significantly to the substrate specificity of DhlA were identified; they include Asp124, Trp125, Phe164, Phe172, Trp175, Phe222, Pro223, and Leu263. These residues are suitable targets for modification by site-directed mutagenesis.

Automated crystal mounting and data collection for protein crystallography.

To increase the efficiency of diffraction data collection for protein crystallographic studies, an automated system designed to store frozen protein crystals, mount them sequentially, align them to the X-ray beam, collect complete data sets, and return the crystals to storage has been developed. Advances in X-ray data collection technology including more brilliant X-ray sources, improved focusing optics, and faster-readout detectors have reduced diffraction data acquisition times from days to hours at a typical protein crystallography laboratory [1,2]. In addition, the number of high-brilliance synchrotron X-ray beam lines dedicated to macromolecular crystallography has increased significantly, and data collection times at these facilities can be routinely less than an hour per crystal. Because the number of protein crystals that may be collected in a 24 hr period has substantially increased, unattended X-ray data acquisition, including automated crystal mounting and alignment, is a desirable goal for protein crystallography. The ability to complete X-ray data collection more efficiently should impact a number of fields, including the emerging structural genomics field [3], structure-directed drug design, and the newly developed screening by X-ray crystallography [4], as well as small molecule applications.

The protein threading problem with sequence amino acid interaction preferences is NP-complete.

In recent protein structure prediction research there has been a great deal of interest in using amino acid interaction preferences (e.g. contact potentials or potentials of mean force) to align ('thread') a protein sequence to a known structural motif. An important open question is whether a polynomial time algorithm for finding the globally optimal threading is possible. We identify the two critical conditions governing this question: (i) variable-length gaps are admitted into the alignment, and (ii) interactions between amino acids from the sequence are admitted into the score function. We prove that if both these conditions are allowed then the protein threading decision problem (does there exist a threading with a score < or = K?) is NP-complete (in the strong sense, i.e. is not merely a number problem) and the related problem of finding the globally optimal protein threading is NP-hard. Therefore, no polynomial time algorithm is possible (unless P = NP). This result augments existing proofs that the direct protein folding problem is NP-complete by providing the corresponding proof for the 'inverse' protein folding problem. It provides a theoretical basis for understanding algorithms currently in use and indicates that computational strategies from other NP-complete problems may be useful for predictive algorithms.