Flow network QSAR for the prediction of physicochemical properties by mapping an electrical resistance network onto a chemical reaction poset.

Usual quantitative structure-activity relationship (QSAR) models are computed from unstructured input data, by using a vector of molecular descriptors for each chemical in the dataset. Another alternative is to consider the structural relationships between the chemical structures, such as molecular similarity, presence of certain substructures, or chemical transformations between compounds. We defined a class of network-QSAR models based on molecular networks induced by a sequence of substitution reactions on a chemical structure that generates a partially ordered set (or poset) oriented graph that may be used to predict various molecular properties with quantitative superstructure-activity relationships (QSSAR). The network-QSAR interpolation models defined on poset graphs, namely average poset, cluster expansion, and spline poset, were tested with success for the prediction of several physicochemical properties for diverse chemicals. We introduce the flow network QSAR, a new poset regression model in which the dataset of chemicals, represented as a reaction poset, is transformed into an oriented network of electrical resistances in which the current flow results in a potential at each node. The molecular property considered in the QSSAR model is represented as the electrical potential, and the value of this potential at a particular node is determined by the electrical resistances assigned to each edge and by a system of batteries. Each node with a known value for the molecular property is attached to a battery that sets the potential on that node to the value of the respective molecular property, and no external battery is attached to nodes from the prediction set, representing chemicals for which the values of the molecular property are not known or are intended to be predicted. The flow network QSAR algorithm determines the values of the molecular property for the prediction set of molecules by applying Ohm's law and Kirchhoff's current law to the poset network of electrical resistances. Several applications of the flow network QSAR are demonstrated.

Chemical graphs, molecular matrices and topological indices in chemoinformatics and quantitative structure-activity relationships.

Chemical and molecular graphs have fundamental applications in chemoinformatics, quantitative structureproperty relationships (QSPR), quantitative structure-activity relationships (QSAR), virtual screening of chemical libraries, and computational drug design. Chemoinformatics applications of graphs include chemical structure representation and coding, database search and retrieval, and physicochemical property prediction. QSPR, QSAR and virtual screening are based on the structure-property principle, which states that the physicochemical and biological properties of chemical compounds can be predicted from their chemical structure. Such structure-property correlations are usually developed from topological indices and fingerprints computed from the molecular graph and from molecular descriptors computed from the three-dimensional chemical structure. We present here a selection of the most important graph descriptors and topological indices, including molecular matrices, graph spectra, spectral moments, graph polynomials, and vertex topological indices. These graph descriptors are used to define several topological indices based on molecular connectivity, graph distance, reciprocal distance, distance-degree, distance-valency, spectra, polynomials, and information theory concepts. The molecular descriptors and topological indices can be developed with a more general approach, based on molecular graph operators, which define a family of graph indices related by a common formula. Graph descriptors and topological indices for molecules containing heteroatoms and multiple bonds are computed with weighting schemes based on atomic properties, such as the atomic number, covalent radius, or electronegativity. The correlation in QSPR and QSAR models can be improved by optimizing some parameters in the formula of topological indices, as demonstrated for structural descriptors based on atomic connectivity and graph distance.

Assessment of 3D models for allergen research.

Allergenic proteins must crosslink specific IgE molecules, bound to the surface of mast cells and basophils, to stimulate an immune response. A structural understanding of the allergen-IgE interface is needed to predict cross-reactivities between allergens and to design hypoallergenic proteins. However, there are less than 90 experimentally determined structures available for the approximately 1500 sequences of allergens and isoallergens cataloged in the Structural Database of Allergenic Proteins. To provide reliable structural data for the remaining proteins, we previously produced more than 500 3D models using an automated procedure, with strict controls on template choice and model quality evaluation. Here, we assessed how well the fold and residue surface exposure of 10 of these models correlated with recently published experimental 3D structures determined by X-ray crystallography or NMR. We also discuss the impact of intrinsically disordered regions on the structural comparison and epitope prediction. Overall, for seven allergens with sequence identities to the original templates higher than 27%, the backbone root-mean square deviations were less than 2 Å between the models and the subsequently determined experimental structures for the ordered regions. Further, the surface exposure of the known IgE epitopes on the models of three major allergens, from peanut (Ara h 1), latex (Hev b 2), and soy (Gly m 4), was very similar to the experimentally determined structures. For the three remaining allergens with lower sequence identities to the modeling templates, the 3D folds were correctly identified. However, the accuracy of those models is not sufficient for a reliable epitope mapping.

AllerML: markup language for allergens.

Many concerns have been raised about the potential allergenicity of novel, recombinant proteins into food crops. Guidelines, proposed by WHO/FAO and EFSA, include the use of bioinformatics screening to assess the risk of potential allergenicity or cross-reactivities of all proteins introduced, for example, to improve nutritional value or promote crop resistance. However, there are no universally accepted standards that can be used to encode data on the biology of allergens to facilitate using data from multiple databases in this screening. Therefore, we developed AllerML a markup language for allergens to assist in the automated exchange of information between databases and in the integration of the bioinformatics tools that are used to investigate allergenicity and cross-reactivity. As proof of concept, AllerML was implemented using the Structural Database of Allergenic Proteins (SDAP; http://fermi.utmb.edu/SDAP/) database. General implementation of AllerML will promote automatic flow of validated data that will aid in allergy research and regulatory analysis.

An Allergen Portrait Gallery: Representative Structures and an Overview of IgE Binding Surfaces.

Recent progress in the biochemical classification and structural determination of allergens and allergen-antibody complexes has enhanced our understanding of the molecular determinants of allergenicity. Databases of allergens and their epitopes have facilitated the clustering of allergens according to their sequences and, more recently, their structures. Groups of similar sequences are identified for allergenic proteins from diverse sources, and all allergens are classified into a limited number of protein structural families. A gallery of experimental structures selected from the protein classes with the largest number of allergens demonstrate the structural diversity of the allergen universe. Further comparison of these structures and identification of areas that are different from innocuous proteins within the same protein family can be used to identify features specific to known allergens. Experimental and computational results related to the determination of IgE binding surfaces and methods to define allergen-specific motifs are highlighted.

Structural analysis of linear and conformational epitopes of allergens.

In many countries regulatory agencies have adopted safety guidelines, based on bioinformatics rules from the WHO/FAO and EFSA recommendations, to prevent potentially allergenic novel foods or agricultural products from reaching consumers. We created the Structural Database of Allergenic Proteins (SDAP, http://fermi.utmb.edu/SDAP/) to combine data that had previously been available only as flat files on Web pages or in the literature. SDAP was designed to be user friendly, to be of maximum use to regulatory agencies, clinicians, as well as to scientists interested in assessing the potential allergenic risk of a protein. We developed methods, unique to SDAP, to compare the physicochemical properties of discrete areas of allergenic proteins to known IgE epitopes. We developed a new similarity measure, the property distance (PD) value that can be used to detect related segments in allergens with clinical observed cross-reactivity. We have now expanded this work to obtain experimental validation of the PD index as a quantitative predictor of IgE cross-reactivity, by designing peptide variants with predetermined PD scores relative to known IgE epitopes. In complementary work we show how sequence motifs characteristic of allergenic proteins in protein families can be used as fingerprints for allergenicity.

The property distance index PD predicts peptides that cross-react with IgE antibodies.

Similarities in the sequence and structure of allergens can explain clinically observed cross-reactivities. Distinguishing sequences that bind IgE in patient sera can be used to identify potentially allergenic protein sequences and aid in the design of hypo-allergenic proteins. The property distance index PD, incorporated in our Structural Database of Allergenic Proteins (SDAP, http://fermi.utmb.edu/SDAP/), may identify potentially cross-reactive segments of proteins, based on their similarity to known IgE epitopes. We sought to obtain experimental validation of the PD index as a quantitative predictor of IgE cross-reactivity, by designing peptide variants with predetermined PD scores relative to three linear IgE epitopes of Jun a 1, the dominant allergen from mountain cedar pollen. For each of the three epitopes, 60 peptides were designed with increasing PD values (decreasing physicochemical similarity) to the starting sequence. The peptides synthesized on a derivatized cellulose membrane were probed with sera from patients who were allergic to Jun a 1, and the experimental data were interpreted with a PD classification method. Peptides with low PD values relative to a given epitope were more likely to bind IgE from the sera than were those with PD values larger than 6. Control sequences, with PD values between 18 and 20 to all the three epitopes, did not bind patient IgE, thus validating our procedure for identifying negative control peptides. The PD index is a statistically validated method to detect discrete regions of proteins that have a high probability of cross-reacting with IgE from allergic patients.

Characteristic motifs for families of allergenic proteins.

The identification of potential allergenic proteins is usually done by scanning a database of allergenic proteins and locating known allergens with a high sequence similarity. However, there is no universally accepted cut-off value for sequence similarity to indicate potential IgE cross-reactivity. Further, overall sequence similarity may be less important than discrete areas of similarity in proteins with homologous structure. To identify such areas, we first classified all allergens and their subdomains in the Structural Database of Allergenic Proteins (SDAP, http://fermi.utmb.edu/SDAP/) to their closest protein families as defined in Pfam, and identified conserved physicochemical property motifs characteristic of each group of sequences. Allergens populate only a small subset of all known Pfam families, as all allergenic proteins in SDAP could be grouped to only 130 (of 9318 total) Pfams, and 31 families contain more than four allergens. Conserved physicochemical property motifs for the aligned sequences of the most populated Pfam families were identified with the PCPMer program suite and catalogued in the webserver MotifMate (http://born.utmb.edu/motifmate/summary.php). We also determined specific motifs for allergenic members of a family that could distinguish them from non-allergenic ones. These allergen specific motifs should be most useful in database searches for potential allergens. We found that sequence motifs unique to the allergens in three families (seed storage proteins, Bet v 1, and tropomyosin) overlap with known IgE epitopes, thus providing evidence that our motif based approach can be used to assess the potential allergenicity of novel proteins.

Weka machine learning for predicting the phospholipidosis inducing potential.

The drug discovery and development process is lengthy and expensive, and bringing a drug to market may take up to 18 years and may cost up to 2 billion $US. The extensive use of computer-assisted drug design techniques may considerably increase the chances of finding valuable drug candidates, thus decreasing the drug discovery time and costs. The most important computational approach is represented by structure-activity relationships that can discriminate between sets of chemicals that are active/inactive towards a certain biological receptor. An adverse effect of some cationic amphiphilic drugs is phospholipidosis that manifests as an intracellular accumulation of phospholipids and formation of concentric lamellar bodies. Here we present structure-activity relationships (SAR) computed with a wide variety of machine learning algorithms trained to identify drugs that have phospholipidosis inducing potential. All SAR models are developed with the machine learning software Weka, and include both classical algorithms, such as k-nearest neighbors and decision trees, as well as recently introduced methods, such as support vector machines and artificial immune systems. The best predictions are obtained with support vector machines, followed by perceptron artificial neural network, logistic regression, and k-nearest neighbors.

Combinatorics of reaction-network posets.

Reaction networks are viewed as derived from ordinary molecular structures related in reactant-product pairs so as to manifest a chemical super-structure. Such super-structures then are candidates for applications in a general combinatoric chemistry. Notable additional characterization of a reaction super-structure occurs when such reaction graphs are directed, as for example when there is progressive substitution (or addition) on a fixed molecular skeleton. Such a set of partially ordered entities is in mathematics termed a poset, which further manifests a number of special properties, as then might be utilized in different applications. Focus on the overall "super-structural" poset goes beyond ordinary molecular structure in attending to how a structure fits into a (reaction) network, and thereby brings an extra "dimension" to conventional stereochemical theory. The possibility that different molecular properties vary smoothly along chains of interconnections in such a super-structure is a natural assumption for a novel approach to molecular property and bioactivity correlations. Different manners to interpolate/extrapolate on a poset network yield quantitative super-structure/activity relationships (QSSARs), with some numerical fits, e.g., for properties of polychlorinated biphenyls (PCBs) seemingly being quite reasonable. There seems to be promise for combinatoric posetic ideas.

Comprehensive 3D-modeling of allergenic proteins and amino acid composition of potential conformational IgE epitopes.

Similarities in sequences and 3D structures of allergenic proteins provide vital clues to identify clinically relevant immunoglobulin E (IgE) cross-reactivities. However, experimental 3D structures are available in the Protein Data Bank for only 5% (45/829) of all allergens catalogued in the Structural Database of Allergenic Proteins (SDAP, http://fermi.utmb.edu/SDAP). Here, an automated procedure was used to prepare 3D-models of all allergens where there was no experimentally determined 3D structure or high identity (95%) to another protein of known 3D structure. After a final selection by quality criteria, 433 reliable 3D models were retained and are available from our SDAP Website. The new 3D models extensively enhance our knowledge of allergen structures. As an example of their use, experimentally derived "continuous IgE epitopes" were mapped on 3 experimentally determined structures and 13 of our 3D-models of allergenic proteins. Large portions of these continuous sequences are not entirely on the surface and therefore cannot interact with IgE or other proteins. Only the surface exposed residues are constituents of "conformational IgE epitopes" which are not in all cases continuous in sequence. The surface exposed parts of the experimental determined continuous IgE epitopes showed a distinct statistical distribution as compared to their presence in typical protein-protein interfaces. The amino acids Ala, Ser, Asn, Gly and particularly Lys have a high propensity to occur in IgE binding sites. The 3D-models will facilitate further analysis of the common properties of IgE binding sites of allergenic proteins.

Robust quantitative modeling of peptide binding affinities for MHC molecules using physical-chemical descriptors.

Major histocompatibility complex (MHC) molecules bind short peptides resulting from intracellular processing of foreign and self proteins, and present them on the cell surface for recognition by T-cell receptors. We propose a new robust approach to quantitatively model the binding affinities of MHC molecules by quantitative structure-activity relationships (QSAR) that use the physical-chemical amino acid descriptors E1-E5. These QSAR models are robust, sequence-based, and can be used as a fast and reliable filter to predict the MHC binding affinity for large protein databases.

Four new topological indices based on the molecular path code.

The sequence of all paths pi of lengths i = 1 to the maximum possible length in a hydrogen-depleted molecular graph (which sequence is also called the molecular path code) contains significant information on the molecular topology, and as such it is a reasonable choice to be selected as the basis of topological indices (TIs). Four new (or five partly new) TIs with progressively improved performance (judged by correctly reflecting branching, centricity, and cyclicity of graphs, ordering of alkanes, and low degeneracy) have been explored. (i) By summing the squares of all numbers in the sequence one obtains Sigmaipi(2), and by dividing this sum by one plus the cyclomatic number, a Quadratic TI is obtained: Q = Sigmaipi(2)/(mu+1). (ii) On summing the Square roots of all numbers in the sequence one obtains Sigmaipi(1/2), and by dividing this sum by one plus the cyclomatic number, the TI denoted by S is obtained: S = Sigmaipi(1/2)/(mu+1). (iii) On dividing terms in this sum by the corresponding topological distances, one obtains the Distance-reduced index D = Sigmai{pi(1/2)/[i(mu+1)]}. Two similar formulas define the next two indices, the first one with no square roots: (iv) distance-Attenuated index: A = Sigmai{pi/[i(mu + 1)]}; and (v) the last TI with two square roots: Path-count index: P = Sigmai{pi(1/2)/[i(1/2)(mu + 1)]}. These five TIs are compared for their degeneracy, ordering of alkanes, and performance in QSPR (for all alkanes with 3-12 carbon atoms and for all possible chemical cyclic or acyclic graphs with 4-6 carbon atoms) in correlations with six physical properties and one chemical property.

Bioinformatics approaches to classifying allergens and predicting cross-reactivity.

Allergenic proteins from very different environmental sources have similar sequences and structures. This fact may account for multiple allergen syndromes, whereby a myriad of diverse plants and foods may induce a similar IgE-based reaction in certain patients. Identifying the common triggering protein in these sources, in silico, can aid designing individualized therapy for allergen sufferers. This article provides an overview of databases on allergenic proteins, and ways to identify common proteins that may be the cause of multiple allergy syndromes. The major emphasis is on the relational Structural Database of Allergenic Proteins (SDAP []), which includes cross-referenced data on the sequence, structure, and IgE epitopes of over 800 allergenic proteins, coupled with specially developed bioinformatics tools to group all allergens and identify discrete areas that may account for cross-reactivity. SDAP is freely available on the Web to clinicians and patients.

Modeling the bioconcentration factors and bioaccumulation factors of polychlorinated biphenyls with posetic quantitative super-structure/activity relationships (QSSAR).

During bioconcentration, chemical pollutants from water are absorbed by aquatic animals via the skin or a respiratory surface, while the entry routes of chemicals during bioaccumulation are both directly from the environment (skin or a respiratory surface) and indirectly from food. The bioconcentration factor (BCF) and the bioaccumulation factor (BAF) for a particular chemical compound are defined as the ratio of the concentration of a chemical inside an organism to the concentration in the surrounding environment. Because the experimental determination of BAF and BCF is time-consuming and expensive, it is efficacious to develop models to provide reliable activity predictions for a large number of chemical compounds. Polychlorinated biphenyls (PCBs) released from industrial activities are persistent pollutants of the environment that produce widespread contamination of water and soil. PCBs can bioaccumulate in the food chain, constituting a potential source of exposure for the general population. To predict the bioconcentration and bioaccumulation factors for PCBs we make use of the biphenyl substitution-reaction network for the sequential substitution of H-atoms by Cl-atoms. Each PCB structure then occurs as a node of this reaction network, which is some sort of super-structure, turning out mathematically to be a partially ordered set (poset). Rather than dealing with the molecular structure via ordinary QSAR we use only this poset, making different quantitative super-structure/activity relationships (QSSAR). Thence we developed cluster expansion and splinoid QSSARs for PCB bioconcentration and bioaccumulation factors. The predictive ability of the BAF and BCF models generated for 20 data sets (representing different conditions and fish species) was evaluated with the leave-one-out cross-validation, which shows that the splinoid QSSAR (r between 0.903 and 0.935) are better than models computed with the cluster expansion (r between 0.745 and 0.887). The splinoid QSSAR models for BAF and BCF yield predictions for the missing PCBs in the investigated data sets.

Common physical-chemical properties correlate with similar structure of the IgE epitopes of peanut allergens.

Although many sequences and linear IgE epitopes of allergenic proteins have been identified and archived in databases, structural and physicochemical discriminators that define their specific properties are lacking. Current bioinformatics tools for predicting the potential allergenicity of a novel protein use methods that were not designed to compare peptides. Novel tools to determine the quantitative sequence and three-dimensional (3D) relationships between IgE epitopes of major allergens from peanut and other foods have been implemented in the Structural Database of Allergenic Proteins (SDAP; http://fermi.utmb.edu/SDAP/). These peptide comparison tools are based on five-dimensional physicochemical property (PCP) vectors. Sequences from SDAP proteins similar in their physicochemical properties to known epitopes of Ara h 1 and Ara h 2 were identified by calculating property distance (PD) values. A 3D model of Ara h 1 was generated to visualize the 3D structure and surface exposure of the epitope regions and peptides with a low PD value to them. Many sequences similar to the known epitopes were identified in related nut allergens, and others were within the sequences of Ara h 1 and Ara h 2. Some of the sequences with low PD values correspond to other known epitopes. Regions with low PD values to one another in Ara h 1 had similar predicted structure, on opposite sides of the internal dimer axis. The PD scale detected epitope pairs that are similar in structure and/or reactivity with patient IgE. The high immunogenicity and IgE reactivity of peanut allergen proteins might be due to the proteins' arrays of similar antigenic regions on opposite sides of a single protein structure.

Posetic quantitative superstructure/activity relationships (QSSARs) for chlorobenzenes.

As a result of the widespread industrial use of polychlorinated hydrocarbons, they have accumulated in nearly all types of environmental compartments, especially in aquatic systems. Particularly, chloroaromatics are among the most undesirable industrial effluents because of their persistence and toxicity. To predict chlorobenzene (CB) toxicities, we make use of a novel scheme that looks beyond simple molecular structure to the manner in which such a structure embeds in an overall reaction network. Thence, a resultant modeling gives a quantitative superstructure/activity relationship (QSSAR) with the (chloro-substitution) reaction network viewed mathematically as a partially ordered set (or poset). Different numerical fittings to the overall poset lead to different QSSAR models, of which we investigate three: average poset, cluster expansion, and splinoid poset QSSAR models for the CBs' toxicities against various species (Poecilia reticulata,Pimephales promelas, Daphnia magna, Rana japonica, etc). Excellent results are obtained for all QSSAR toxicity models. On the basis of the poset reaction diagram, all three of these QSSAR models reflect, in distinct ways, the topology of the network that describes the interconversion of chemical species. Although in the majority of investigated datasets all poset QSSAR models give very good predictions, in some cases, they complement each other. These differences show that more reliable predictions can be obtained by using a consensus prediction that combines data from the three posetic models.