Neural Upscaling from Residue-Level Protein Structure Networks to Atomistic Structures.

Coarse-graining is a powerful tool for extending the reach of dynamic models of proteins and other biological macromolecules. Topological coarse-graining, in which biomolecules or sets thereof are represented via graph structures, is a particularly useful way of obtaining highly compressed representations of molecular structures, and simulations operating via such representations can achieve substantial computational savings. A drawback of coarse-graining, however, is the loss of atomistic detail-an effect that is especially acute for topological representations such as protein structure networks (PSNs). Here, we introduce an approach based on a combination of machine learning and physically-guided refinement for inferring atomic coordinates from PSNs. This "neural upscaling" procedure exploits the constraints implied by PSNs on possible configurations, as well as differences in the likelihood of observing different configurations with the same PSN. Using a 1 µs atomistic molecular dynamics trajectory of Aß1-40, we show that neural upscaling is able to effectively recapitulate detailed structural information for intrinsically disordered proteins, being particularly successful in recovering features such as transient secondary structure. These results suggest that scalable network-based models for protein structure and dynamics may be used in settings where atomistic detail is desired, with upscaling employed to impute atomic coordinates from PSNs.

Protein structure networks provide insight into active site flexibility in esterase/lipases from the carnivorous plant Drosera capensis.

In plants, esterase/lipases perform transesterification reactions, playing an important role in the synthesis of useful molecules, such as those comprising the waxy coatings of leaf surfaces. Plant genomes and transcriptomes have provided a wealth of data about expression patterns and the circumstances under which these enzymes are upregulated, e.g. pathogen defense and response to drought; however, predicting their functional characteristics from genomic or transcriptome data is challenging due to weak sequence conservation among the diverse members of this group. Although functional sequence blocks mediating enzyme activity have been identified, progress to date has been hampered by the paucity of information on the structural relationships among these regions and how they affect substrate specificity. Here we present methodology for predicting overall protein flexibility and active site flexibility based on molecular modeling and analysis of protein structure networks (PSNs). We define two new types of specialized PSNs: sequence region networks (SRNs) and active site networks (ASNs), which provide parsimonious representations of molecular structure in reference to known features of interest. Our approach, intended as an aid to target selection for poorly characterized enzyme classes, is demonstrated for 26 previously uncharacterized esterase/lipases from the genome of the carnivorous plant Drosera capensis and validated using a case/control design. Analysis of the network relationships among functional blocks and among the chemical moieties making up the catalytic triad reveals potentially functionally significant differences that are not apparent from sequence analysis alone.

Computational Studies of Intrinsically Disordered Proteins.

Frequently elusive to experimental characterizations, intrinsically disordered proteins (IDPs) can be probed using molecular dynamics to provide detailed insight into their complex structure, dynamics, and function. However, previous computational studies were often found to disagree with experiment due to either force field biases or insufficient sampling. In this study, nine unstructured short peptides and the HIV-1 Rev protein were simulated and extended to microseconds to assess these limitations in IDP simulations. In short peptide simulations, a tested IDP-specific force field ff14IDPSFF outperforms its generic counterpart ff14SB as agreement of simulated NMR observables with experiment improves, though its advantages are not clear-cut in apo Rev simulations. It is worth noting that sampling is probably still not sufficient in the ff14SB simulations of apo Rev even if 10 ms have been collected. This indicates that enhanced sampling techniques would greatly benefit IDP simulations. Finally, detailed structural analyses of apo Rev conformations demonstrate different secondary structural preferences between ff14SB (helical) and ff14IDPSFF (random coil). A natural next step is to ask a more quantitative question: whether ff14SB is too ordered or ff14IDPSFF is too disordered in simulations of more complex IDPs such as Rev. This requires further quantitative analyses both experimentally and computationally.

Structure prediction and network analysis of chitinases from the Cape sundew, Drosera capensis.

BACKGROUND: Carnivorous plants possess diverse sets of enzymes with novel functionalities applicable to biotechnology, proteomics, and bioanalytical research. Chitinases constitute an important class of such enzymes, with future applications including human-safe antifungal agents and pesticides. Here, we compare chitinases from the genome of the carnivorous plant Drosera capensis to those from related carnivorous plants and model organisms. METHODS: Using comparative modeling, in silico maturation, and molecular dynamics simulation, we produce models of the mature enzymes in aqueous solution. We utilize network analytic techniques to identify similarities and differences in chitinase topology. RESULTS: Here, we report molecular models and functional predictions from protein structure networks for eleven new chitinases from D. capensis, including a novel class IV chitinase with two active domains. This architecture has previously been observed in microorganisms but not in plants. We use a combination of comparative and de novo structure prediction followed by molecular dynamics simulation to produce models of the mature forms of these proteins in aqueous solution. Protein structure network analysis of these and other plant chitinases reveal characteristic features of the two major chitinase families. GENERAL SIGNIFICANCE: This work demonstrates how computational techniques can facilitate quickly moving from raw sequence data to refined structural models and comparative analysis, and to select promising candidates for subsequent biochemical characterization. This capability is increasingly important given the large and growing body of data from high-throughput genome sequencing, which makes experimental characterization of every target impractical.