Ri-Bo Huang

Prediction of protein structural classes using support vector machines

Summary.The support vector machine, a machine-learning method, is used to predict the four structural classes, i.e. mainly α, mainly β, α–β and fss, from the topology-level of CATH protein structure database. For the binary classification, any two structural classes which do not share any secondary structure such as α and β elements could be classified with as high as 90% accuracy. The accuracy, however, will decrease to less than 70% if the structural classes to be classified contain structure elements in common. Our study also shows that the dimensions of feature space 202 = 400 (for dipeptide) and 203 = 8 000 (for tripeptide) give nearly the same prediction accuracy. Among these 4 structural classes, multi-class classification gives an overall accuracy of about 52%, indicating that the multi-class classification technique in support of vector machines may still need to be further improved in future investigation.

The multiple roles of histidine in protein interactions.

BackgroundAmong the 20 natural amino acids histidine is the most active and versatile member that plays the multiple roles in protein interactions, often the key residue in enzyme catalytic reactions. A theoretical and comprehensive study on the structural features and interaction properties of histidine is certainly helpful.ResultsFour interaction types of histidine are quantitatively calculated, including: (1) Cation-π interactions, in which the histidine acts as the aromatic π-motif in neutral form (His), or plays the cation role in protonated form (His+); (2) π-π stacking interactions between histidine and other aromatic amino acids; (3) Hydrogen-π interactions between histidine and other aromatic amino acids; (4) Coordinate interactions between histidine and metallic cations. The energies of π-π stacking interactions and hydrogen-π interactions are calculated using CCSD/6-31+G(d,p). The energies of cation-π interactions and coordinate interactions are calculated using B3LYP/6-31+G(d,p) method and adjusted by empirical method for dispersion energy.ConclusionsThe coordinate interactions between histidine and metallic cations are the strongest one acting in broad range, followed by the cation-π, hydrogen-π, and π-π stacking interactions. When the histidine is in neutral form, the cation-π interactions are attractive; when it is protonated (His+), the interactions turn to repulsive. The two protonation forms (and pKa values) of histidine are reversibly switched by the attractive and repulsive cation-π interactions. In proteins the π-π stacking interaction between neutral histidine and aromatic amino acids (Phe, Tyr, Trp) are in the range from -3.0 to -4.0 kcal/mol, significantly larger than the van der Waals energies.

Insights from investigating the interaction of oseltamivir (Tamiflu) with neuraminidase of the 2009 H1N1 swine flu virus.

The neuraminidase (NA) of influenza virus is the target of anti-flu drugs oseltamivir and zanamivir. Clinical practices showed that oseltamivir was effective to treat the 2009-H1N1 influenza but failed to the 2006-H5N1 avian influenza. To perform an in-depth analysis on such a drug-resistance problem, the 2009-H1N1-NA structure was developed. To compare it with the crystal 2006-H5N1-NA structure as well as the 1918 influenza virus H1N1-NA structure, the multiple sequential and structural alignments were performed. It has been revealed that the hydrophobic residue Try347 in H5N1-NA does not match with the hydrophilic carboxyl group of oseltamivir as in the case of H1N1-NA. This may be the reason why H5N1 avian influenza virus is drug-resistant to oseltamivir. The finding provides useful insights for how to modify the existing drugs, such as oseltamivir and zanamivir, making them not only become more effective against H1N1 virus but also effective against H5N1 virus.

Energetic analysis of the two controversial drug binding sites of the M2 proton channel in influenza A virus.

Understanding the mechanism of the M2 proton channel of influenza A is crucially important to both basic research and drug discovery. Recently, the structure was determined independently by high-resolution NMR and X-ray crystallography. However, the two studies lead to completely different drug-binding mechanisms: the X-ray structure shows the drug blocking the pore from inside; whereas the NMR structure shows the drug inhibiting the channel from outside by an allosteric mechanism. Which one of the two is correct? To address this problem, we conducted an in-depth computational analysis. The conclusions drawn from various aspects, such as energetics, the channel-gating dynamic process, the pK(a) shift and its impact on the channel, and the consistency with the previous functional studies, among others, are all in favour to the allosteric mechanism revealed by the NMR structure. The findings reported here may stimulate and encourage new strategies for developing effective drugs against influenza A, particularly in dealing with the drug-resistant problems.

Multiple field three dimensional quantitative structure–activity relationship (MF-3D-QSAR)

A new drug design method, the multiple field three‐dimensional quantitative structure–activity relationship (MF‐3D‐QSAR), is proposed. It is a combination and development of classical 2D‐QSAR and traditional 3D‐QSAR. In addition to the electrostatic and van der Waals potentials, more potential fields (such as lipophilic potential, hydrogen bonding potential, and nonthermodynamic factors) are integrated in the MF‐3D‐QSAR. Meanwhile, a principal component analysis (PCA) and iterative double least square (IDLS) technique is developed for predicting the bioactivity of query drug candidates. As an example, the MF‐3D‐QSAR is applied to the design of neuraminidase inhibitor and to prove its predictive power, and some useful findings are obtained for developing drugs against influenza virus.

Designing Inhibitors of M2 Proton Channel against H1N1 Swine Influenza Virus

Background M2 proton channel of H1N1 influenza A virus is the target protein of anti-flu drugs amantadine and rimantadine. However, the two once powerful adamantane-based drugs lost their 90% bioactivity because of mutations of virus in recent twenty years. The NMR structure of the M2 channel protein determined by Schnell and Chou (Nature, 2008, 451, 591–595) may help people to solve the drug-resistant problem and develop more powerful new drugs against H1N1 influenza virus. Methodology Docking calculation is performed to build the complex structure between receptor M2 proton channel and ligands, including existing drugs amantadine and rimantadine, and two newly designed inhibitors. The computer-aided drug design methods are used to calculate the binding free energies, with the computational biology techniques to analyze the interactions between M2 proton channel and adamantine-based inhibitors. Conclusions 1) The NMR structure of M2 proton channel provides a reliable structural basis for rational drug design against influenza virus. 2) The channel gating mechanism and the inhibiting mechanism of M2 proton channel, revealed by the NMR structure of M2 proton channel, provides the new ideas for channel inhibitor design. 3) The newly designed adamantane-based inhibitors based on the modeled structure of H1N1-M2 proton channel have two pharmacophore groups, which act like a “barrel hoop”, holding two adjacent helices of the H1N1-M2 tetramer through the two pharmacophore groups outside the channel. 4) The inhibitors with such binding mechanism may overcome the drug resistance problem of influenza A virus to the adamantane-based drugs.

Fragment-based quantitative structure–activity relationship (FB-QSAR) for fragment-based drug design

In cooperation with the fragment‐based design a new drug design method, the so‐called “fragment‐based quantitative structure–activity relationship” (FB‐QSAR) is proposed. The essence of the new method is that the molecular framework in a family of drug candidates are divided into several fragments according to their substitutes being investigated. The bioactivities of molecules are correlated with the physicochemical properties of the molecular fragments through two sets of coefficients in the linear free energy equations. One coefficient set is for the physicochemical properties and the other for the weight factors of the molecular fragments. Meanwhile, an iterative double least square (IDLS) technique is developed to solve the two sets of coefficients in a training data set alternately and iteratively. The IDLS technique is a feedback procedure with machine learning ability. The standard Two‐dimensional quantitative structure–activity relationship (2D‐QSAR) is a special case, in the FB‐QSAR, when the whole molecule is treated as one entity. The FB‐QSAR approach can remarkably enhance the predictive power and provide more structural insights into rational drug design. As an example, the FB‐QSAR is applied to build a predictive model of neuraminidase inhibitors for drug development against H5N1 influenza virus.

Peptide reagent design based on physical and chemical properties of amino acid residues.

It has tremendous values for both drug discovery and basic research to develop a solid bioinformatical tool for guiding peptide reagent design. Based on the physical and chemical properties of amino acids, a new strategy for peptide reagent design, the so‐called AABPD (amino acid based‐peptide design), is proposed. The peptide samples in a training dataset are described by a series of HMLP (heuristic molecular lipophilicity potential) parameters and other physicochemical properties of amino acid residues that form a three‐dimensional data matrix where each component is defined by three indexes: the first index refers to the peptide samples, the second to the amino acid positions, and the third to the amino acid parameters. The binding free energy between a peptide ligand and its protein receptor is calculated by a linear free energy equation through the physicochemical parameters, resulting in a set of simultaneous linear equations between the bioactivity of the peptides and the physicochemical properties of amino acids. An iterative double least square technique is developed for the solution of the three‐dimensional simultaneous linear equation set to determine the amino acid position coefficients of peptide sequence and the physicochemical parameter coefficients of amino acid residues alternately. The two sets of coefficients thus obtained are used for predicting the bioactivity of other query peptide reagents. Two calculation examples, the peptide substrate specificity of the SARS coronavirus 3C‐like proteinase and the affinity prediction for epitope‐peptides with Class I MHC molecules are studied by using the peptide reagent design strategy.

Saturation-mutagenesis in two positions distant from active site of a Klebsiella pneumoniae glycerol dehydratase identifies some highly active mutants

Synthesis of 1,3-propanediol (1,3-PD) from glycerol through the biotransformation process requires two steps, catalyzed by glycerol dehydratase (GDHt) and 1,3-PD oxidoreductase. GDHt is the rate-limiting enzyme in this process. All recombinant microorganisms for production of 1,3-PD so far utilized the natural genes that may not have been optimized. Two positions, which are 19.3A and 29.6A away from the active site in GDHt from Klebsiella pneumoniae, were subjected to saturation-mutagenesis and 38 mutants were characterized. The catalytic activity of a mutant in beta-subunit (beta-Q42F, 29.6A from the active site) was 8.3-fold higher than the wild type, and the enzyme efficiency of other two mutants beta-Q42L and beta-Q42S for substrate glycerol was 336-fold and 80-fold higher than that for 1,2-propanediol. This investigation supplied further evidence that distant mutations could be a good source of diversity and therefore, made a contribution to the toolbox of industrial enzyme improvement.

Energies and physicochemical properties of cation–π interactions in biological structures

The cation-π interactions occur frequently within or between proteins due to six (Phe, Tyr, Trp, Arg, Lys, and His) of the twenty natural amino acids potentially interacting with metallic cations via these interactions. In this study, quantum chemical calculations and molecular orbital (MO) theory are used to study the energies and properties of cation-π interactions in biological structures. The cation-π interactions of H⁺ and Li⁺ are similar to hydrogen bonds and lithium bonds, respectively, in which the small, naked cations H⁺ and Li⁺ are buried deep within the π-electron density of aromatic molecules, forming stable cation-π bonds that are much stronger than the cation-π interactions of other alkali metal cations. The cation-π interactions of metallic cations with atomic masses greater than that of Li⁺ arise mainly from the coordinate bond comprising empty valence atomic orbitals (AOs) of metallic cations and π-MOs of aromatic molecules, though electrostatic interactions may also contribute to the cation-π interaction. The binding strength of cation-π interactions is determined by the charge and types of AOs in the metallic cations. Cation-π interaction energies are distance- and orientation-dependent; energies decrease with the distance (r) and the orientation angle (θ). In solution, the cation-π energies decrease with the increase of the dielectric constant (ɛ) of the solvent; however, solvation has less influence on the H⁺-π and H₃O⁺-π interactions than on interactions with other cations. The conclusions from this study provide useful theoretical insights into the nature of cation-π interactions and may contribute to the development of better force field parameters for describing the molecular dynamics of cation-π interactions within and between proteins.