Hai-Feng Chen

Comparative study of QSAR/QSPR correlations using support vector machines, radial basis function neural networks, and multiple linear regression.

Support vector machines (SVMs) were used to develop QSAR models that correlate molecular structures to their toxicity and bioactivities. The performance and predictive ability of SVM are investigated and compared with other methods such as multiple linear regression and radial basis function neural network methods. In the present study, two different data sets were evaluated. The first one involves an application of SVM to the development of a QSAR model for the prediction of toxicities of 153 phenols, and the second investigation deals with the QSAR model between the structures and the activities of a set of 85 cyclooxygenase 2 (COX-2) inhibitors. For each application, the molecular structures were described using either the physicochemical parameters or molecular descriptors. In both studied cases, the predictive ability of the SVM model is comparable or superior to those obtained by MLR and RBFNN. The results indicate that SVM can be used as an alternative powerful modeling tool for QSAR studies.

ASD: a comprehensive database of allosteric proteins and modulators.

Allostery is the most direct, rapid and efficient way of regulating protein function, ranging from the control of metabolic mechanisms to signal-transduction pathways. However, an enormous amount of unsystematic allostery information has deterred scientists who could benefit from this field. Here, we present the AlloSteric Database (ASD), the first online database that provides a central resource for the display, search and analysis of structure, function and related annotation for allosteric molecules. Currently, ASD contains 336 allosteric proteins from 101 species and 8095 modulators in three categories (activators, inhibitors and regulators). Proteins are annotated with a detailed description of allostery, biological process and related diseases, and modulators with binding affinity, physicochemical properties and therapeutic area. Integrating the information of allosteric proteins in ASD should allow for the identification of specific allosteric sites of a given subtype among proteins of the same family that can potentially serve as ideal targets for experimental validation. In addition, modulators curated in ASD can be used to investigate potent allosteric targets for the query compound, and also help chemists to implement structure modifications for novel allosteric drug design. Therefore, ASD could be a platform and a starting point for biologists and medicinal chemists for furthering allosteric research. ASD is freely available at http://mdl.shsmu.edu.cn/ASD/.

Computational Approach to Drug Design for Oxazolidinones as Antibacterial Agents

A three dimensional Quantitative Structure Activity Relationship (3D-QSAR) model for a series of (S)-3-Aryl-5-substituted oxazolidinones was developed to gain insights into the design for potential new antibacterial agents. It was found that the Comparative Molecular Field Analysis (CoMFA) method yielded good results while the Comparative Molecular Similarity Indices Analysis (CoMSIA) was less satisfactory. The CoMFA method yielded a cross-validated correlation coefficient q(2) = 0.681, non-cross-validated R(2) = 0.991, SE (Standard Error ) = 0.054, and the value of statistical significance measure F = 266.98. The relative steric and electrostatic contributions are 0.542 and 0.458, respectively. These results indicate that the model possesses a high predictivity. Guided by this model, three new compounds were synthesized. All these compounds exhibit inhibitory activity; two of them were shown having high activity (MIC = 1.0 microg/ml). The activity observed by experiments was in good agreement with the theoretical one. It is anticipated that the present model would be of value in facilitating design of new potent antibacterial agents.

New Force Field on Modeling Intrinsically Disordered Proteins

Intrinsically disordered proteins or intrinsically disordered protein regions comprise a large portion of eukaryotic proteomes (between 35% and 51%). These intrinsically disordered proteins were found to link with cancer and various other diseases. However, widely used additive force field parameter sets are insufficient in quantifying the structural properties of intrinsically disordered proteins. Therefore, we explored to a systematic correction of a base additive force field parameter set (chosen as Amber ff99SBildn) to correct the biases that was first demonstrated in simulations with the base parameter set. Specifically, the φ/ψ distributions of disorder‐promoting residues were systematically corrected with the CMAP method. Our simulations show that the CMAP corrected Amber parameter set, termed ff99IDPs, improves the φ/ψ distributions of the disorder‐promoting residues with respect to the benchmark data of intrinsically disordered protein structures, with root mean‐squared percentage deviation less than 0.15% between the simulation and the benchmark. Our further validation shows that the chemical shifts from the ff99IDPs simulations are in quantitative agreement with those from reported NMR measurements for two tested IDPs, MeV NTAIL, and p53. The predicted residue dipolar couplings also show high correlation with experimental data. Interestingly, our simulations show that ff99IDPs can still be used to model the ordered state when the intrinsically disordered proteins are in complex, in contrast to ff99SBildn that can be applied well only to the ordered complex structures. These findings confirm that the newly proposed Amber ff99IDPs parameter set provides a reasonable tool in further studies of intrinsically disordered protein structures. In addition, our study also shows the importance of considering intrinsically disordered protein structures in general‐purposed force field developments for both additive and non‐additive models.

Molecular dynamics simulation of phosphorylated KID post-translational modification.

Background Kinase-inducible domain (KID) as transcriptional activator can stimulate target gene expression in signal transduction by associating with KID interacting domain (KIX). NMR spectra suggest that apo-KID is an unstructured protein. After post-translational modification by phosphorylation, KID undergoes a transition from disordered to well folded protein upon binding to KIX. However, the mechanism of folding coupled to binding is poorly understood. Methodology To get an insight into the mechanism, we have performed ten trajectories of explicit-solvent molecular dynamics (MD) for both bound and apo phosphorylated KID (pKID). Ten MD simulations are sufficient to capture the average properties in the protein folding and unfolding. Conclusions Room-temperature MD simulations suggest that pKID becomes more rigid and stable upon the KIX-binding. Kinetic analysis of high-temperature MD simulations shows that bound pKID and apo-pKID unfold via a three-state and a two-state process, respectively. Both kinetics and free energy landscape analyses indicate that bound pKID folds in the order of KIX access, initiation of pKID tertiary folding, folding of helix αB, folding of helix αA, completion of pKID tertiary folding, and finalization of pKID-KIX binding. Our data show that the folding pathways of apo-pKID are different from the bound state: the foldings of helices αA and αB are swapped. Here we also show that Asn139, Asp140 and Leu141 with large Φ-values are key residues in the folding of bound pKID. Our results are in good agreement with NMR experimental observations and provide significant insight into the general mechanisms of binding induced protein folding and other conformational adjustment in post-translational modification.

Induced fit or conformational selection for RNA/U1A folding

The hairpin II of U1 snRNA can bind U1A protein with high affinity and specificity. NMR spectra suggest that the loop region of apo-RNA is largely unstructured and undergoes a transition from unstructured to well-folded upon U1Abinding. However, the mechanism that RNA folding coupled protein binding is poorly understood. To get an insight into the mechanism, we have performed explicit-solvent molecular dynamics (MD) to study the folding kinetics of bound RNA and apo-RNA. Room-temperature MD simulations suggest that the conformation of bound RNA has significant adjustment and becomes more stable upon U1A binding. Kinetic analysis of high-temperature MD simulations shows that bound RNA and apo-RNA unfold via a two-state process, respectively. Both kinetics and free energy landscape analyses indicate that bound RNA folds in the order of RNA contracting, U1A binding, and tertiary folding. The predicted Phi-values suggest that A8, C10, A11, and G16 are key bases for bound RNA folding. Mutant Arg52Gln analysis shows that electrostatic interaction and hydrogen bonds between RNA and U1A (Arg52Gln) decrease. These results are in qualitative agreement with experiments. Furthermore, this method could be used in other studies about biomolecule folding upon receptor binding.

Enhancing the Promiscuous Phosphotriesterase Activity of a Thermostable Lactonase (GkaP) for the Efficient Degradation of Organophosphate Pesticides

ABSTRACT The phosphotriesterase-like lactonase (PLL) enzymes in the amidohydrolase superfamily hydrolyze various lactones and exhibit latent phosphotriesterase activities. These enzymes serve as attractive templates for in vitro evolution of neurotoxic organophosphates (OPs) with hydrolytic capabilities that can be used as bioremediation tools. Here, a thermostable PLL from Geobacillus kaustophilus HTA426 (GkaP) was targeted for joint laboratory evolution with the aim of enhancing its catalytic efficiency against OP pesticides. By a combination of site saturation mutagenesis and whole-gene error-prone PCR approaches, several improved variants were isolated. The most active variant, 26A8C, accumulated eight amino acid substitutions and demonstrated a 232-fold improvement over the wild-type enzyme in reactivity (k cat/Km ) for the OP pesticide ethyl-paraoxon. Concomitantly, this variant showed a 767-fold decrease in lactonase activity with δ-decanolactone, imparting a specificity switch of 1.8 × 105-fold. 26A8C also exhibited high hydrolytic activities (19- to 497-fold) for several OP pesticides, including parathion, diazinon, and chlorpyrifos. Analysis of the mutagenesis sites on the GkaP structure revealed that most mutations are located in loop 8, which determines substrate specificity in the amidohydrolase superfamily. Molecular dynamics simulation shed light on why 26A8C lost its native lactonase activity and improved the promiscuous phosphotriesterase activity. These results permit us to obtain further insights into the divergent evolution of promiscuous enzymes and suggest that laboratory evolution of GkaP may lead to potential biological solutions for the efficient decontamination of neurotoxic OP compounds.

Mechanism of Coupled Folding and Binding in the siRNA-PAZ Complex.

The PAZ domain plays a key role in gene silencing pathway. The PAZ domain binds with siRNAs to form the multimeric RNA-induced silencing complex (RISC). RISC identifies mRNAs homologous to the siRNAs and promotes their degradation. It was found that binding with siRNA significantly enhances apo-PAZ folding. However, the mechanism by which folding is coupled to binding is poorly understood. Thus, the coupling relationship between binding and folding is very important for understanding the function of gene silencing. We have performed molecular dynamics (MD) of both bound and apo-PAZ to study the coupling mechanism between binding and folding in the siRNA-PAZ complex. Room-temperature MD simulations suggest that both PAZ and siRNA become more rigid and stable upon siRNA binding. Kinetic analysis of high-temperature MD simulations shows that both bound and apo-PAZ unfold via a two-state process. The unfolding pathways are different between bound and apo-PAZ: the order of helix III and helices I & II unfolding is switched. Furthermore, transition probability was used to determine the transition state ensemble for both bound and apo-PAZ. It was found that the transition state of bound PAZ is more compact than that of apo-PAZ. The predicted Φ-values suggest that the Φ-values of helix III and sheets of β3-β7 for bound PAZ are more native-like than those of apo-PAZ upon the binding of siRNA. The results can help us to understand the mechanism of gene silencing.

Quantitative predictions of gas chromatography retention indexes with support vector machines, radial basis neural networks and multiple linear regression

Support vector machines (SVM), radial basis function neural networks (RBFNN) and multiple linear regression (MLR) methods were used to investigate the correlation between GC retention indexes (RI) and physicochemical descriptors for both 174 and 132 diverse organic compounds. The correlation coefficient r(2) between experimental and predicted retention index for training and test sets by SVM, RBFNN and MLR is 0.986, 0.976 and 0.971 (for 174 compounds), 0.986, 0.951 and 0.963 (for 132 compounds) respectively. The results show that non-linear SVM derives statistical models have similar prediction ability to those of RBFNN and MLR methods. This indicates that SVM can be used as an alternative modeling tool for quantitative structure-property/activity relationship (QSPR/QSAR) studies.

All-Atom Computer Simulations of Amyloid Fibrils Disaggregation

Amyloidlike fibrils are found in many fatal diseases, including Alzheimers disease, type II diabetes mellitus, transmissible spongiform encephalopathies, and prion diseases. These diseases are linked to proteins that have partially unfolded, misfolded, and aggregated into amyloidlike fibrils. The kinetics of amyloidlike fibrils aggregation is still hotly debated and remains an important open question. We have utilized the GNNQQNY crystal structure and high-temperature molecular dynamics simulation in explicit solvent to study the disaggregation mechanism of the GNNQQNY fibrils and to infer its likely aggregation pathways. A hexamer model and a 12-mer model both with two parallel beta-sheets separated by a dry side-chain interface were adopted in our computational analysis. A cumulative time of 1 micros was simulated for the hexamer model at five different temperatures (298 K, 348 K, 398 K, 448 K, and 498 K), and a cumulative time of 2.1 micros was simulated for the 12-mer model at four temperatures (298 K, 398 K, 448 K, and 498 K). Our disaggregation landscape and kinetics analyses indicate that tetramers probably act as the transition state in both the hexamer and the 12-mer simulations. In addition, the 12-mer simulations show that the initial aggregation nucleus is with eight peptides. Furthermore, the landscape is rather flat from 8-mers to 12-mers, indicating the absence of major barriers once the initial aggregation nucleus forms. Thus, the likely aggregation pathway is from monomers to the initial nucleus of 8-mers with tetramers as the transition state. Transition state structure analysis shows that the two dominant transition state conformations are tetramers in the 3-1 and 2-2 arrangements. The predominant nucleus conformations are in peptide arrangements maximizing dry side-chain contacts. Landscape and kinetics analyses also indicate that the parallel beta-sheets form earlier than the dry side-chain contacts during aggregation. These results provide further insights in understanding the early fibrils aggregation.