A top-down method for the determination of residue-specific solvent accessibility in proteins

We present a method employing top-down Fourier transform mass spectrometry (FTMS) for the rapid profiling of amino acid side-chain reactivity. The reactivity of side-chain groups can be used to infer residue-specific solvent accessibility and can also be used in the same way as H/D exchange reactions to probe protein structure and interactions. We probed the reactivity of the N-terminal and epsilon-lysine amino groups of ubiquitin by reaction with N-hydroxysuccinimidyl acetate (NHSAc), which specifically acetylates primary amines. Using a hybrid Q-FTMS instrument, we observed several series of multiply acetylated ubiquitin ions that varied with the NHSAc:protein stoichiometry. We isolated and fragmented each member of the series of acetylated ubiquitin ions in the front end of the instrument and measured the fragment ion masses in the FTMS analyzer cell to determine which residue positions were modified. As we increased the NHSAc:protein stoichiometric ratio, identification of the fragments from native protein and protein with successively increasing modification allowed the assignment of the complete order of reactivity of the primary amino groups in ubiquitin (Met 1 approximately Lys 6 approximately Lys 48 approximately Lys 63>Lys 33>Lys 11>Lys 27, Lys 29). These results are in excellent agreement with the reactivity expected from other studies and predicted from the known crystal structure of ubiquitin. The top-down approach eliminates the need for proteolytic digestion, high-performance liquid chromatographic separations and all other chemical steps except the labeling reaction, making it rapid and amenable to automation using small quantities of protein.     

Dimensionality of amino acid space and solvent accessibility prediction with neural networks

Solvent accessibility prediction from amino acid sequences has been pursued by several researchers. Such a prediction typically starts by transforming the amino acid category (or type) information into numerical representations. All twenty amino acids can be completely and uniquely represented by 20-dimensional vectors. Here, we investigate if the amino acid space defined in this way really requires twenty dimensions. We tried to develop corresponding representations in fewer dimensions. A method for searching optimal codification schema in an arbitrary space using neural networks was developed. The method is used to obtain optimal encoding of amino acids at various levels of dimensionality, and applied to optimize the amino acid codifications for the prediction of the solvent accessibility values of the proteins using feed-forward neural networks. The traditional 20-dimensional codification seems to be redundant in solving the solvent accessibility prediction problem, since a 1-dimensional codification is able to achieve almost the same degree of accuracy as the 20-dimensional codification. Optimal coding in much fewer dimensions could be used to make the predictions of accessible surface area with almost the same degree of accuracy as that obtained by a fully unique 20-dimensional coding. The 1-dimensional amino acid codification for solvent accessibility prediction obtained by a purely mathematical way based on neural networks is highly correlated with a physical property of the amino acids, namely their average solvent accessibility. The method developed to find the optimal codification is general, although the codification thus produced is dependent on the type of estimated property.     

A consensus procedure improving solvent accessibility prediction

Prediction methods of structural features in 1D represent a useful tool for the understanding of folding, classification, and function of proteins, and, in particular, for 3D structure prediction. Among the structural aspects characterizing a protein, solvent accessibility has received great attention in recent years. The available methods proposed for predicting accessibility have never considered the combination of the results deriving from different methods to construct a consensus prediction able to provide more reliable results. A consensus approach that increases prediction accuracy using three high-performance methods is described. The results of our method for three different protein data sets show that up to 3.0% improvement in prediction accuracy of solvent accessibility may be obtained by a consensus approach. The improvement also extends to the correlation coefficient. Application of our consensus approach to the accessibility prediction using only three prediction methods gives results better than single methods combined for consensus formation. Currently, the scarce availability of predictors with similar parameters defining solvent accessibility hinders the testing of other methods in our consensus procedure.     

Prediction of solvent accessibility of amino acid residues: critical aspects

For predicting solvent accessibility from the sequence of amino acids in proteins, we use a logistic function trained on a non-redundant protein database. Using a principal component analysis, we find that the prediction can be considered, in a good approximation, as a monofactorial problem: a crossed effect of the burial propensity of amino acids and of their locations at positions flanking the amino acid of interest. Complementary effects depend on the presence of certain amino acids (mostly P, G and C) at given positions. We have refined the predictive model (1) by adding supplementary input data, (2) by using a strategy of prediction correction and (3) by adapting the decision rules according to the amino acid type. We obtain a best score of 77.6% correct prediction for a relative accessibility of 9%. However, compared to trivial strategy only based upon the frequencies of buried or exposed residues, the gain is less than 4%. Received: 4 June 1998 / Accepted: 17 September 1998 / Published online: 10 December 1998     

A semi-implicit solvent model for the simulation of peptides and proteins

We present a new model of biomolecules hydration based on macroscopic electrostatic theory, that can both describe the microscopic details of solvent-solute interactions and allow for an efficient evaluation of the electrostatic hydration free energy. This semi-implicit model considers the solvent as an ensemble of polarizable pseudoparticles whose induced dipole describe both the electronic and orientational solvent polarization. In the presented version of the model, there is no mutual dipolar interaction between the particles, and they only interact through short-ranged Lennard-Jones interactions. The model has been integrated into a molecular dynamics code, and offers the possibility to simulate efficiently the conformational evolution of biomolecules. It is able to provide estimations of the electrostatic solvation free energy within short time windows during the simulation. It has been applied to the study of two small peptides, the octaalanine and the N-terminal helix of ribonuclease A, and two proteins, the bovine pancreatic trypsin inhibitor and the B1 immunoglobin-binding domain of streptococcal protein G. Molecular dynamics simulations of these biomolecules, using a slightly modified Amber force field, provide stable and meaningful trajectories in overall agreement with experiments and all-atom simulations. Correlations with respect to Poisson-Boltzmann electrostatic solvation free energies are also presented to discuss the parameterization of the model and its consequences.     

DeepSIRT: A deep neural network for identification of sirtuin targets and their subcellular localizations

Sirtuins are a family of proteins that play a key role in regulating a wide range of cellular processes including DNA regulation, metabolism, aging/longevity, cell survival, apoptosis, and stress resistance. Sirtuins are protein deacetylases and include in the class III family of histone deacetylase enzymes (HDACs). The class III HDACs contains seven members of the sirtuin family from SIRT1 to SIRT7. The seven members of the sirtuin family have various substrates and are present in nearly all subcellular localizations including the nucleus, cytoplasm, and mitochondria. In this study, a deep neural network approach using one-dimensional Convolutional Neural Networks (CNN) was proposed to build a prediction model that can accurately identify the outcome of the sirtuin protein by targeting their subcellular localizations. Therefore, the function and localization of sirtuin targets were analyzed and annotated to compartmentalize into distinct subcellular localizations. We further reduced the sequence similarity between protein sequences and three feature extraction methods were applied in datasets. Finally, the proposed method has been tested and compared with various machine-learning algorithms. The proposed method is validated on two independent datasets and showed an average of up to 85.77 % sensitivity, 97.32 % specificity, and 0.82 MCC for seven members of the sirtuin family of proteins.     

A polarizable force field for water using an artificial neural network

A force field for liquid water including polarization effects has been constructed using an artificial neural network (ANN). It is essential to include a many-body polarization effect explicitly into a potential energy function in order to treat liquid water which is dense and highly polar. The new potential energy function is a combination of empirical and nonempirical potentials. The TIP4P model was used for the empirical part of the potential. For the nonempirical part, an ANN with a back-propagation of error algorithm (BPNN) was introduced to reproduce the complicated many-body interaction energy surface from ab initio quantum mechanical calculations. BPNN, described in terms of a matrix, provides enough flexibility to describe the complex potential energy surface (PES). The structural and thermodynamic properties, calculated by isobaric-isothermal (constant-NPT) Monte Carlo simulations with the new polarizable force field for water, are compatible with experimental results. Thus, the simulation establishes the validity of using our estimated PES with a polarization effect for accurate predictions of liquid state properties. Applications of this approach are simple and systematic so that it can easily be applied to the development of other force fields besides the water-water system.     

Combinative neural network and its applications

A new approach named combinative neural network (CN) using partial least squares (PLS) analysis to modify the hidden layer in the multi-layered feed forward (MLFF) neural networks (NN) was proposed in this paper. The significant contributions of PLS in the proposed CN are to reorganize the outputs of hidden nodes such that the correlation of variables could be circumvented, to make the network meet the non-linear relationship best between the input and output data of the NN, and to eliminate the risk of over-fitting problem at the same time. The performance of the proposed approach was demonstrated through two examples, a well defined nonlinear approximation problem, and a practical nonlinear pattern classification problem with unknown relationship between the input and output data. The results were compared with those by conventional MLFF NNs. Good performance and time-saving implementation make the proposed method an attractive approach for non-linear mapping and classification.     

Elucidating protein secondary structure with circular dichroism and a neural network

Circular dichroism spectroscopy is a quick method for determining the average secondary structures of proteins, probing their interactions with their environment, and aiding drug discovery. This article describes the development of a self‐organising map structure‐fitting methodology named secondary structure neural network (SSNN) to aid this process and reduce the level of expertise required. SSNN uses a database of spectra from proteins with known X‐ray structures; prediction of structures for new proteins is then possible. It has been designed as 3 units: SSNN1 takes spectra for known proteins, clusters them into a map, and SSNN2 creates a matching structure map. SSNN3 places unknown spectra on the map and gives them structure vectors. SSNN3 output illustrates the process and results obtained. We detail the strengths and weaknesses of SSNN and compare it with widely accepted structure fitting programs. Current input format is Δ? per amino acid residue from 240 to 190 nm in 1 nm steps for the known and unknown proteins and a vector summarizing the secondary structure elements of the known proteins. The format is readily modified to include input data with, for example, extended wavelength ranges or different assignment of secondary structures. SSNN can be used either pretrained with a reference set from the CDPro web site (direct application of SSNN3, with the provided output from SSNN1 and SSNN2) or all three modules can be used as required. SSNN3 is available trained (with the reference set of the 48‐spectra set used in this work complemented by five additional spectra) at http://www2.warwick.ac.uk/fac/sci/chemistry/research/arodger/arodgergroup/research_intro/instrumentation/ssnn/ . © 2013 Wiley Periodicals, Inc.     

Application of different training methodologies for the development of a back propagation artificial neural network retention model in ion chromatography

The reliability of predicted separations in ion chromatography depends mainly on the accuracy of retention predictions. Any model able to improve this accuracy will yield predicted optimal separations closer to the reality. In this work artificial neural networks were used for retention modeling of void peak, fluoride, chlorite, chloride, chlorate, nitrate and sulfate. In order to increase performance characteristics of the developed model, different training methodologies were applied and discussed. Furthermore, the number of neurons in hidden layer, activation function and number of experimental data used for building the model were optimized in terms of decreasing the experimental effort without disruption of performance characteristics. This resulted in the superior predictive ability of developed retention model (average of relative error is 0.4533%). Copyright © 2008 John Wiley & Sons, Ltd.     

Optimization of injection‐molding process for mechanical properties of polypropylene components via a generalized regression neural network

This study analyzed contour distortions, wear and tensile properties of polypropylene (PP) components applied in the interior coffer of automobiles. A hybrid method integrating a trained generalized regression neural network (GRNN) and a sequential quadratic programming (SQP) method is proposed to determine an optimal parameter setting of the injection‐molding process. The specimens were prepared under different injection‐molding conditions by changing melting temperatures, injection speeds, and injection pressures. Average contour distortions at six critical locations, wear and tensile properties were selected as the quality targets. Sixteen experimental runs, based on a Taguchi orthogonal array table, were utilized to train the GRNN and then the SQP method was applied to search for an optimal setting. The trained GRNN was capable of predicting average contour distortions, wear and tensile properties at various injection‐molding conditions. In addition, the analysis of variance (ANOVA) was implemented to identify significant factors for the molding process and the proposed algorithm was compared with traditional schemes like the Taguchi method and the design of experiments (DOE) approach. Copyright © 2007 John Wiley & Sons, Ltd.     

Behaviour of proteins in alternate solvent conditions

Recent progress in thermodynamic aspects of protein function in unusual environments are described. Systems considered include enzymes incorporated in reverse micelles, immobilized onto solid support or suspended in low-water solvents.

---

Zusammenfassung Es werden die bisherigen Ergebnisse bei der Untersuchung thermodynamischer Aspekte von Proteinfunktionen in ungewöhnlicher Umgebung beschrieben. Untersuchte Systeme beinhalten Enzyme, die in reversen, auf fester Unterlage immobilisierten oder in Lösungsmitteln suspendierten Mizellen eingeschlossen wurden.

     

Experimental study on the 3D‐printed plastic parts and predicting the mechanical properties using artificial neural networks

This study investigates the mechanical properties of 3D‐printed plastic parts fabricated using Fused Deposition Modeling (FDM). For this purpose, a 3D printer named KASAME was designed and built by the researchers. The test samples were fabricated using polylactic acid (PLA). The experiments were conducted using three melt temperatures (190°C, 205°C, and 220°C), four layer thickness values (0.06 mm, 0.10 mm, 0.19 mm, and 0.35 mm), and three raster pattern orientations (+45°/?45° [the crisscross pattern], horizontal and vertical). Tensile strength tests were performed to determine tensile strength values of the samples and fracture surfaces were also analyzed. Using artificial neural networks, a mathematical model for the tensile test results was generated corresponding to the raster pattern employed in 3D fabrication. Tensile strength tests indicated that melt temperature, layer thickness, and raster pattern orientation had a significant effect on the tensile strengths of the samples. According to the result of the experiment, the maximum average tensile strength values were observed for the samples fabricated using the crisscross raster pattern. The analysis of variance (ANOVA) table shows the raster pattern (PCR) value of 48.68% was obtained with the highest degree of influence. With respect to R ², the best performing artificial neural network model, with test and training values of 0.999199 and 0.999997, respectively, was observed to be the crisscross raster pattern. Copyright © 2016 John Wiley & Sons, Ltd.     

Optimization of extraction of linarin from Flos chrysanthemi indici by response surface methodology and artificial neural network

The extraction of linarin from Flos chrysanthemi indici by ethanol was investigated. Two modeling techniques, response surface methodology and artificial neural network, were adopted to optimize the process parameters, such as, ethanol concentration, extraction period, extraction frequency, and solvent to material ratio. We showed that both methods provided good predictions, but artificial neural network provided a better and more accurate result. The optimum process parameters include, ethanol concentration of 74%, extraction period of 2 h, extraction three times, solvent to material ratio of 12 mL/g. The experiment yield of linarin was 90.5% that deviated less than 1.6% from that obtained by predicted result.     

Microdroplet size prediction in microfluidic systems via artificial neural network modeling for water-in-oil emulsion formulation

In this paper, an experimental study and modeling by artificial neural networks were carried out to predict the generated microdroplet dimensionless size in a microfluidic system in order to formulate a water-in-oil emulsion. The various parameters that affect the size of microdroplets (flow rates, viscosities, surface tensions of both the two phases and the diameter of the microchannel) are studied and further grouped into dimensionless numbers; we used these numbers as input to the neural network and the dimensionless length as output. The better neural network architecture has 10 neurons in the hidden layer with a mean square error of 1.4 10^?6 and a determination’s coefficient near 1 value. The relative importance of inputs on the size of the microdroplets has been determined using the Garson algorithm and the results are in good agreement with other works.     

Artificial neural networks applied for studying metallic complexes

Metallic complexes of multimetal and multiligand systems are complicated for calculating equilibrium concentrations in solutions. An artificial neural network has been developed for studying Al³⁺ and EDTA complexes in solution with an initial concentration of 0.01 mol L^?1 for these species. In this system there are 20 compounds and may exist 18 simultaneous reactions. The neural network has been trained and the simulated data of different concentrations as a function of pH are predicted with an accuracy of about 1% for all species simultaneously. A general analytical formula is presented, which directly relates all the concentrations as a function of pH. The analysis showed that predictions closer to the boundary of the input and output data are quantitative while out of these limits these are not even qualitative. © 2001 John Wiley & Sons, Inc. J Comput Chem 22: 1691–1701, 2001     

A comparison of classifiers for predicting the class color of fluorescent proteins

Fluorescent proteins have been applied in a wide variety of fields ranging from basic science to industrial applications. Apart from the naturally occurring fluorescent proteins, there is a growing interest in genetically modified variants that emit light in a specific wavelength. Genetically modifying a protein is not an easy task, especially because the exchange of one residue by other has to achieve the desired property while maintaining protein stability. To help in the choice of residue exchange, computational methods are applied to predict function and stability of proteins. In this work we have prepared a dataset composed by 109 fluorescent proteins and tested four classical supervised classification algorithms: artificial neural networks (ANNs), decision trees (DTs), support vector machines (SVMs) and random forests (RFs). This is the first time that algorithms are compared in this task. Results of comparing the algorithm's performance shows that DT, SVM and RF were significantly better than ANNs, and RF was the best method in all the scenarios. However, the interpretability of DTs is highly relevant and can provide important clues about the mechanisms involved in protein color emission. The results are promising and indicate that the use of in silico methods can greatly reduce the time and cost of the in vitro experiments.     

A neural network approach to infrared spectrum interpretation

The simple linear neural network model was investigated as a method for automated interpretation of infrared spectra. The model was trained using a database of infrared spectra of organic compounds of known structure. The model was able to learn, without any prior input of spectrum-structure correlations, to recognize and identify 76 functional groupings with accuracies ranging from fair to excellent. The effect of network input parameters and of training set composition were studied, and several sources of spurious correlations were identified and corrected.Dedicated to Professor W. Simon on the occasion of his 60th birthday