Scoring docking models with evolutionary information

We have developed methods for the extraction of evolutionary information from multiple sequence alignments for use in the study of the evolution of protein interaction networks and in the prediction of protein interaction. For Rounds 3, 4, and 5 of the CAPRI experiment, we used scores derived from the analysis of multiple sequence alignments to submit predictions for 7 of the 12 targets. Our docking models were generated with Hex and GRAMM, but all our predictions were selected using methods based on multiple sequence alignments and on the available experimental evidence. With this approach, we were able to predict acceptable level models for 4 of the targets, and for a fifth target, we located the residues involved in the binding surface. Here we detail our successes and highlight several of the limitations and problems that we faced while dealing with particular docking cases.     

Individual‐based ecology of coastal birds

Conservation objectives for non‐breeding coastal birds (shorebirds and wildfowl) are determined from their population size at coastal sites. To advise coastal managers, models must predict quantitatively the effects of environmental change on population size or the demographic rates (mortality and reproduction) that determine it. As habitat association models and depletion models are not able to do this, we developed an approach that has produced such predictions thereby enabling policy makers to make evidence‐based decisions. Our conceptual framework is individual‐based ecology, in which populations are viewed as having properties (e.g. size) that arise from the traits (e.g. behaviour, physiology) and interactions of their constituent individuals. The link between individuals and populations is made through individual‐based models (IBMs) that follow the fitness‐maximising decisions of individuals and predict population‐level consequences (e.g. mortality rate) from the fates of these individuals. Our first IBM was for oystercatchers Haematopus ostralegus and accurately predicted their density‐dependent mortality. Subsequently, IBMs were developed for several shorebird and wildfowl species at several European sites, and were shown to predict accurately overwinter mortality, and the foraging behaviour from which predictions are derived. They have been used to predict the effect on survival in coastal birds of sea level rise, habitat loss, wind farm development, shellfishing and human disturbance. This review emphasises the wider applicability of the approach, and identifies other systems to which it could be applied. We view the IBM approach as a very useful contribution to the general problem of how to advance ecology to the point where we can routinely make meaningful predictions of how populations respond to environmental change.     

ZPRED: predicting the distance to the membrane center for residues in alpha-helical membrane proteins

MOTIVATION: Prediction methods are of great importance for membrane proteins as experimental information is harder to obtain than for globular proteins. As more membrane protein structures are solved it is clear that topology information only provides a simplified picture of a membrane protein. Here, we describe a novel challenge for the prediction of alpha-helical membrane proteins: to predict the distance between a residue and the center of the membrane, a measure we define as the Z-coordinate. Even though the traditional way of depicting membrane protein topology is useful, it is advantageous to have a measure that is based on a more "physical" property such as the Z-coordinate, since it implicitly contains information about re-entrant helices, interfacial helices, the tilt of a transmembrane helix and loop lengths. RESULTS: We show that the Z-coordinate can be predicted using either artificial neural networks, hidden Markov models or combinations of both. The best method, ZPRED, uses the output from a hidden Markov model together with a neural network. The average error of ZPRED is 2.55A and 68.6% of the residues are predicted within 3A of the target Z-coordinate in the 5-25A region. ZPRED is also able to predict the maximum protrusion of a loop to within 3A for 78% of the loops in the dataset. AVAILABILITY: Supplementary information and training data is available at http://www.sbc.su.se/~erikgr/.     

Selective prediction-set models with coverage rate guarantees

The current approach to using machine learning (ML) algorithms in healthcare is to either require clinician oversight for every use case or use their predictions without any human oversight. We explore a middle ground that lets ML algorithms abstain from making a prediction to simultaneously improve their reliability and reduce the burden placed on human experts. To this end, we present a general penalized loss minimization framework for training selective prediction-set (SPS) models, which choose to either output a prediction set or abstain. The resulting models abstain when the outcome is difficult to predict accurately, such as on subjects who are too different from the training data, and achieve higher accuracy on those they do give predictions for. We then introduce a model-agnostic, statistical inference procedure for the coverage rate of an SPS model that ensembles individual models trained using K-fold cross-validation. We find that SPS ensembles attain prediction-set coverage rates closer to the nominal level and have narrower confidence intervals for its marginal coverage rate. We apply our method to train neural networks that abstain more for out-of-sample images on the MNIST digit prediction task and achieve higher predictive accuracy for ICU patients compared to existing approaches.     

Candidate epitope identification using peptide property models: application to cancer immunotherapy

Peptides derived from pathogens or tumors are selectively presented by the major histocompatibility complex proteins (MHC) to the T lymphocytes. Antigenic peptide-MHC complexes on the cell surface are specifically recognized by T cells and, in conjunction with co-factor interactions, can activate the T cells to initiate the necessary immune response against the target cells. Peptides that are capable of binding to multiple MHC molecules are potential T cell epitopes for diverse human populations that may be useful in vaccine design. Bioinformatical approaches to predict MHC binding peptides can facilitate the resource-consuming effort of T cell epitope identification. We describe a new method for predicting MHC binding based on peptide property models constructed using biophysical parameters of the constituent amino acids and a training set of known binders. The models can be applied to development of anti-tumor vaccines by scanning proteins over-expressed in cancer cells for peptides that bind to a variety of MHC molecules. The complete algorithm is described and illustrated in the context of identifying candidate T cell epitopes for melanomas and breast cancers. We analyzed MART-1, S-100, MBP, and CD63 for melanoma and p53, MUC1, cyclin B1, HER-2/neu, and CEA for breast cancer. In general, proteins over-expressed in cancer cells may be identified using DNA microarray expression profiling. Comparisons of model predictions with available experimental data were assessed. The candidate epitopes identified by such a computational approach must be evaluated experimentally but the approach can provide an efficient and focused strategy for anti-cancer immunotherapy development.     

Global mapping of pharmacological space

We present the global mapping of pharmacological space by the integration of several vast sources of medicinal chemistry structure-activity relationships (SAR) data. Our comprehensive mapping of pharmacological space enables us to identify confidently the human targets for which chemical tools and drugs have been discovered to date. The integration of SAR data from diverse sources by unique canonical chemical structure, protein sequence and disease indication enables the construction of a ligand-target matrix to explore the global relationships between chemical structure and biological targets. Using the data matrix, we are able to catalog the links between proteins in chemical space as a polypharmacology interaction network. We demonstrate that probabilistic models can be used to predict pharmacology from a large knowledge base. The relationships between proteins, chemical structures and drug-like properties provide a framework for developing a probabilistic approach to drug discovery that can be exploited to increase research productivity.     

Predicting Growth Conditions from Internal Metabolic Fluxes in an In-Silico Model of E. coli

A widely studied problem in systems biology is to predict bacterial phenotype from growth conditions, using mechanistic models such as flux balance analysis (FBA). However, the inverse prediction of growth conditions from phenotype is rarely considered. Here we develop a computational framework to carry out this inverse prediction on a computational model of bacterial metabolism. We use FBA to calculate bacterial phenotypes from growth conditions in E. coli, and then we assess how accurately we can predict the original growth conditions from the phenotypes. Prediction is carried out via regularized multinomial regression. Our analysis provides several important physiological and statistical insights. First, we show that by analyzing metabolic end products we can consistently predict growth conditions. Second, prediction is reliable even in the presence of small amounts of impurities. Third, flux through a relatively small number of reactions per growth source (∼10) is sufficient for accurate prediction. Fourth, combining the predictions from two separate models, one trained only on carbon sources and one only on nitrogen sources, performs better than models trained to perform joint prediction. Finally, that separate predictions perform better than a more sophisticated joint prediction scheme suggests that carbon and nitrogen utilization pathways, despite jointly affecting cellular growth, may be fairly decoupled in terms of their dependence on specific assortments of molecular precursors.     

Sources and predictors of resolvable indel polymorphism assessed using rice as a model

The principal sources of genetic variation that can be assayed with restriction enzymes are base substitutions and insertions/deletions (indels). The likelihood of detecting indels as restriction fragment length polymorphisms (RFLPs) is determined by the size and frequency of the indels, and the ability to resolve small indels as RFLPs is limited by the distribution of restriction fragment sizes. In this study, we use aligned sequences from the indica and japonica subspecies of rice ( Oryza sativa L.) to quantify and compare the ability of restriction enzymes to detect indels. We look specifically at two abundant transposable element-derived indel sources: miniature inverted repeat transposable elements (MITEs) and long terminal repeat (LTR) retroelements. From this analysis we conclude that indels rather than base substitutions are the prevailing source of the polymorphism detected in rice. We show that, although MITE derived indels are more abundant than LTR-retroelement derived indels, LTR-retroelements have a greater capacity to generate visible restriction fragment length polymorphism because of their larger size. We find that the variation in the detectability of indels among restriction enzymes can be explained by differences in the frequency and dispersion of their restriction sites in the genome. The parameters that describe the fragment size distributions obtained with the restriction enzymes are highly correlated across the sequenced genomes of rice, Arabidopsis and human, with the exception of some extreme deviations in frequency for particular recognition sequences corresponding to variations in the levels and modes of DNA methylation in the three disparate organisms. Thus, we can predict the relative ability of a restriction enzyme to detect indels derived from a specific source based on the distribution of restriction fragment sizes, even when this is estimated for a distantly related genome.Electronic Supplementary Material Supplementary Material is available in the online version of this article at Communicated by M.-A. Grandbastien     

Persistent random motion: Uncovering cell migration dynamics

In this paper we study analytically the stick-slip models recently introduced to explain the stochastic migration of free cells. We show that persistent motion of cells of many different types is compatible with stochastic reorientation models which admit an analytical mesoscopic treatment. This is proved by examining and discussing experimental data compiled from different sources in the literature, and by fitting some of these results too. We are able to explain many of the ‘apparently complex’ migration patterns obtained recently from cell tracking data, like power-law dependences in the mean square displacement or non-Gaussian behavior for the kurtosis and the velocity distributions, which depart from the predictions of the classical Ornstein-Uhlenbeck process.     

Computational prediction of host-pathogen protein-protein interactions

MOTIVATION: Infectious diseases such as malaria result in millions of deaths each year. An important aspect of any host-pathogen system is the mechanism by which a pathogen can infect its host. One method of infection is via protein-protein interactions (PPIs) where pathogen proteins target host proteins. Developing computational methods that identify which PPIs enable a pathogen to infect a host has great implications in identifying potential targets for therapeutics. RESULTS: We present a method that integrates known intra-species PPIs with protein-domain profiles to predict PPIs between host and pathogen proteins. Given a set of intra-species PPIs, we identify the functional domains in each of the interacting proteins. For every pair of functional domains, we use Bayesian statistics to assess the probability that two proteins with that pair of domains will interact. We apply our method to the Homo sapiens-Plasmodium falciparum host-pathogen system. Our system predicts 516 PPIs between proteins from these two organisms. We show that pairs of human proteins we predict to interact with the same Plasmodium protein are close to each other in the human PPI network and that Plasmodium pairs predicted to interact with same human protein are co-expressed in DNA microarray datasets measured during various stages of the Plasmodium life cycle. Finally, we identify functionally enriched sub-networks spanned by the predicted interactions and discuss the plausibility of our predictions. AVAILABILITY: Supplementary data are available at http://staff.vbi.vt.edu/dyermd/publications/dyer2007a.html. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.     

Functional evaluation of domain-domain interactions and human protein interaction networks

MOTIVATION: Large amounts of protein and domain interaction data are being produced by experimental high-throughput techniques and computational approaches. To gain insight into the value of the provided data, we used our new similarity measure based on the Gene Ontology (GO) to evaluate the molecular functions and biological processes of interacting proteins or domains. The applied measure particularly addresses the frequent annotation of proteins or domains with multiple GO terms. RESULTS: Using our similarity measure, we compare predicted domain-domain and human protein-protein interactions with experimentally derived interactions. The results show that our similarity measure is of significant benefit in quality assessment and confidence ranking of domain and protein networks. We also derive useful confidence score thresholds for dividing domain interaction predictions into subsets of low and high confidence. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.     

The influence of villous counter current exchange on intestinal absorption.

The vasculature of an intestinal villus in the cat consists of one arterial limb and several venous limbs connected hairpin-like. The distance between the arterial and venous limbs (length about 1 mm) amounts to 10–20 μm so that the basic requirements for an extravascular shunt and a counter current exchange (CCE) are fulfilled. Equations are derived which describe the influence of a CCE in the intestinal villi on intestinal absorption and elimination, respectively. A CCE retards the absorption or elimination, diminishes the absorption or elimination rate, and builds up a gradient from tip to base during absorption or basis to tip during elimination. The efficiency of the CCE is increased with decreasing villous blood flow (especially with decreasing linear flow rate) and increasing permeability of the substances. The hitherto published experimental data obtained in the cat correspond with the theoretical predictions. The lack of experimental evidence for CCE in the rabbit and rat small intestine may be due to the different structure of their villous vasculature.     

Predicting the compliance of small diameter vascular grafts from uniaxial tensile tests

The compliance hypothesis states that the compliance of vascular grafts should match that of the host artery for optimal patency. Although this has not been proven, the literature shows that much effort has gone into measuring compliance. Uniaxial circumferential tensile tests are simpler than compliance tests, but do not give the compliance (a multiaxial property) directly. Therefore, we have used mechanical models to correlate the two. Simple models suffer from inappropriate simplifying assumptions. In a clinically useful range, a Laplace law model and an incremental elasticity model do not predict the compliance from the rigidity as well as does a model derived from finite elasticity. This latter model has helped locate sources of errors. Variations in graft thickness, diameter, and anisotropy may be responsible for scatter in the experimental correlations between compliance and uniaxial rigidity.     

Accurate quantification of functional analogy among close homologs

Correctly evaluating functional similarities among homologous proteins is necessary for accurate transfer of experimental knowledge from one organism to another, and is of particular importance for the development of animal models of human disease. While the fact that sequence similarity implies functional similarity is a fundamental paradigm of molecular biology, sequence comparison does not directly assess the extent to which two proteins participate in the same biological processes, and has limited utility for analyzing families with several parologous members. Nevertheless, we show that it is possible to provide a cross-organism functional similarity measure in an unbiased way through the exclusive use of high-throughput gene-expression data. Our methodology is based on probabilistic cross-species mapping of functionally analogous proteins based on Bayesian integrative analysis of gene expression compendia. We demonstrate that even among closely related genes, our method is able to predict functionally analogous homolog pairs better than relying on sequence comparison alone. We also demonstrate that the landscape of functional similarity is often complex and that definitive "functional orthologs" do not always exist. Even in these cases, our method and the online interface we provide are designed to allow detailed exploration of sources of inferred functional similarity that can be evaluated by the user.     

Modeling of the structure of bacteriorhodopsin. A molecular dynamics study.

Secondary structure predictions for membrane proteins are relatively reliable and permit the construction of model structures that may serve as initial conformations for molecular dynamics simulations. This might provide a scheme to predict the three-dimensional structures of membrane proteins. The feasibility of such an approach is tested for bacteriorhodopsin. We were not able to fully predict the kidney-shaped structure of bacteriorhodopsin. However, features compatible with this structure developed in a simulation starting from a circular arrangement of the seven predicted helices. When instead we started from the kidney shape, assigning the seven predicted helices in different ways to those on the structure, we could distinguish between the different assignments on the basis of energy and tilt of the helices. In this way we could select the correct assignment from a few others. For the correct assignment, the helices spontaneously adopted a tilt that agrees remarkably well with the experimental model structure derived by others. The root-mean-square deviation between our best molecular dynamics structure and the experimental model structure is 3.8 A, caused mainly by deviations in the internal degrees of freedom of the helices.     

A molecular dynamics strategy for CSαβ peptides disulfide-assisted model refinement

Many cysteine-stabilized antimicrobial peptides from a variety of living organisms could be good candidates for the development of anti-infective agents. In the absence of experimentally obtained structural data, peptide modeling is an essential tool for understanding structure–activity relationships and for optimizing the bioactive moieties. Focusing on cysteine-rich peptide structures, we reproduced the case of structure predictions in the so-called midnight zone. We developed our protocol on a training set derived by clustering the available cysteine-stabilized αβ (CSαβ) structures in nine different representative families and tested it on peptides randomly selected from each family. Starting from draft models, we tested a structure-based disulfide predictor and we used cysteine distances as constraints during molecular dynamics. Finally, we proposed an analysis for final structure selection. Accordingly, we obtained a mean root mean square deviation improvement of 21% for the test set. Our findings demonstrate that it is possible to predict the network of disulfide bridges in cysteine-stabilized peptides and to use this result to improve the accuracy of structural predictions. Finally, we applied the methods to predict the structure of royalisin, a cysteine-rich peptide with unknown structure.