Cristian R. Munteanu

Automatic feature extraction using genetic programming: An application to epileptic EEG classification

This paper applies genetic programming (GP) to perform automatic feature extraction from original feature database with the aim of improving the discriminatory performance of a classifier and reducing the input feature dimensionality at the same time. The tree structure of GP naturally represents the features, and a new function generated in this work automatically decides the number of the features extracted. In experiments on two common epileptic EEG detection problems, the classification accuracy on the GP-based features is significant higher than on the original features. Simultaneously, the dimension of the input features for the classifier is much smaller than that of the original features.

MIND-BEST: Web Server for Drugs and Target Discovery; Design, Synthesis, and Assay of MAO-B Inhibitors and Theoretical−Experimental Study of G3PDH Protein from Trichomonas gallinae

Many drugs with very different affinity to a large number of receptors are described. Thus, in this work, we selected drug-target pairs (DTPs/nDTPs) of drugs with high affinity/nonaffinity for different targets. Quantitative structure-activity relationship (QSAR) models become a very useful tool in this context because they substantially reduce time and resource-consuming experiments. Unfortunately, most QSAR models predict activity against only one protein target and/or they have not been implemented on a public Web server yet, freely available online to the scientific community. To solve this problem, we developed a multitarget QSAR (mt-QSAR) classifier combining the MARCH-INSIDE software for the calculation of the structural parameters of drug and target with the linear discriminant analysis (LDA) method in order to seek the best model. The accuracy of the best LDA model was 94.4% (3,859/4,086 cases) for training and 94.9% (1,909/2,012 cases) for the external validation series. In addition, we implemented the model into the Web portal Bio-AIMS as an online server entitled MARCH-INSIDE Nested Drug-Bank Exploration & Screening Tool (MIND-BEST), located at http://miaja.tic.udc.es/Bio-AIMS/MIND-BEST.php . This online tool is based on PHP/HTML/Python and MARCH-INSIDE routines. Finally, we illustrated two practical uses of this server with two different experiments. In experiment 1, we report for the first time a MIND-BEST prediction, synthesis, characterization, and MAO-A and MAO-B pharmacological assay of eight rasagiline derivatives, promising for anti-Parkinson drug design. In experiment 2, we report sampling, parasite culture, sample preparation, 2-DE, MALDI-TOF and -TOF/TOF MS, MASCOT search, 3D structure modeling with LOMETS, and MIND-BEST prediction for different peptides as new protein of the found in the proteome of the bird parasite Trichomonas gallinae, which is promising for antiparasite drug targets discovery.

Multi-target QPDR classification model for human breast and colon cancer-related proteins using star graph topological indices

Abstract The cancer diagnostic is a complex process and, sometimes, the specific markers can interfere or produce negative results. Thus, new simple and fast theoretical models are required. One option is the complex network graphs theory that permits us to describe any real system, from the small molecules to the complex genetic, neural or social networks by transforming real properties in topological indices. This work converts the protein primary structure data in specific Randics star networks topological indices using the new sequence to star networks (S2SNet) application. A set of 1054 proteins were selected from previous works and contains proteins related or not with two types of cancer, human breast cancer (HBC) and human colon cancer (HCC). The general discriminant analysis method generates an input-coded multi-target classification model with the training/predicting set accuracies of 90.0% for the forward stepwise model type. In addition, a protein subset was modified by single amino acid mutations with higher log-odds PAM250 values and tested with the new classification if can be related with HBC or HCC. In conclusion, we shown that, using simple input data such is the primary protein sequence and the simples linear analysis, it is possible to obtain accurate classification models that can predict if a new protein related with two types of cancer. These results promote the use of the S2SNet in clinical proteomics.

Trypano-PPI: a web server for prediction of unique targets in trypanosome proteome by using electrostatic parameters of protein-protein interactions.

Trypanosoma brucei causes African trypanosomiasis in humans (HAT or African sleeping sickness) and Nagana in cattle. The disease threatens over 60 million people and uncounted numbers of cattle in 36 countries of sub-Saharan Africa and has a devastating impact on human health and the economy. On the other hand, Trypanosoma cruzi is responsible in South America for Chagas disease, which can cause acute illness and death, especially in young children. In this context, the discovery of novel drug targets in Trypanosome proteome is a major focus for the scientific community. Recently, many researchers have spent important efforts on the study of protein-protein interactions (PPIs) in pathogen Trypanosome species concluding that the low sequence identities between some parasite proteins and their human host render these PPIs as highly promising drug targets. To the best of our knowledge, there are no general models to predict Unique PPIs in Trypanosome (TPPIs). On the other hand, the 3D structure of an increasing number of Trypanosome proteins is reported in databases. In this regard, the introduction of a new model to predict TPPIs from the 3D structure of proteins involved in PPI is very important. For this purpose, we introduced new protein-protein complex invariants based on the Markov average electrostatic potential xi(k)(R(i)) for amino acids located in different regions (R(i)) of i-th protein and placed at a distance k one from each other. We calculated more than 30 different types of parameters for 7866 pairs of proteins (1023 TPPIs and 6823 non-TPPIs) from more than 20 organisms, including parasites and human or cattle hosts. We found a very simple linear model that predicts above 90% of TPPIs and non-TPPIs both in training and independent test subsets using only two parameters. The parameters were (d)xi(k)(s) = |xi(k)(s(1)) - xi(k)(s(2))|, the absolute difference between the xi(k)(s(i)) values on the surface of the two proteins of the pairs. We also tested nonlinear ANN models for comparison purposes but the linear model gives the best results. We implemented this predictor in the web server named TrypanoPPI freely available to public at http://miaja.tic.udc.es/Bio-AIMS/TrypanoPPI.php. This is the first model that predicts how unique a protein-protein complex in Trypanosome proteome is with respect to other parasites and hosts, opening new opportunities for antitrypanosome drug target discovery.

Enzymes/non-enzymes classification model complexity based on composition, sequence, 3D and topological indices

The huge amount of new proteins that need a fast enzymatic activity characterization creates demands of protein QSAR theoretical models. The protein parameters that can be used for an enzyme/non-enzyme classification includes the simpler indices such as composition, sequence and connectivity, also called topological indices (TIs) and the computationally expensive 3D descriptors. A comparison of the 3D versus lower dimension indices has not been reported with respect to the power of discrimination of proteins according to enzyme action. A set of 966 proteins (enzymes and non-enzymes) whose structural characteristics are provided by PDB/DSSP files was analyzed with Python/Biopython scripts, STATISTICA and Weka. The list of indices includes, but it is not restricted to pure composition indices (residue fractions), DSSP secondary structure protein composition and 3D indices (surface and access). We also used mixed indices such as composition-sequence indices (Chous pseudo-amino acid compositions or coupling numbers), 3D-composition (surface fractions) and DSSP secondary structure amino acid composition/propensities (obtained with our Prot-2S Web tool). In addition, we extend and test for the first time several classic TIs for the Randics protein sequence Star graphs using our Sequence to Star Graph (S2SG) Python application. All the indices were processed with general discriminant analysis models (GDA), neural networks (NN) and machine learning (ML) methods and the results are presented versus complexity, average of Shannons information entropy (Sh) and data/method type. This study compares for the first time all these classes of indices to assess the ratios between model accuracy and indices/model complexity in enzyme/non-enzyme discrimination. The use of different methods and complexity of data shows that one cannot establish a direct relation between the complexity and the accuracy of the model.

3D entropy and moments prediction of enzyme classes and experimental-theoretic study of peptide fingerprints in Leishmania parasites.

The number of protein 3D structures without function annotation in Protein Data Bank (PDB) has been steadily increased. This fact has led in turn to an increment of demand for theoretical models to give a quick characterization of these proteins. In this work, we present a new and fast Markov chain model (MCM) to predict the enzyme classification (EC) number. We used both linear discriminant analysis (LDA) and/or artificial neural networks (ANN) in order to compare linear vs. non-linear classifiers. The LDA model found is very simple (three variables) and at the same time is able to predict the first EC number with an overall accuracy of 79% for a data set of 4755 proteins (859 enzymes and 3896 non-enzymes) divided into both training and external validation series. In addition, the best non-linear ANN model is notably more complex but has an overall accuracy of 98.85%. It is important to emphasize that this method may help us to predict not only new enzyme proteins but also to select peptide candidates found on the peptide mass fingerprints (PMFs) of new proteins that may improve enzyme activity. In order to illustrate the use of the model in this regard, we first report the 2D electrophoresis (2DE) and MADLI-TOF mass spectra characterization of the PMF of a new possible malate dehydrogenase sequence from Leishmania infantum. Next, we used the models to predict the contribution to a specific enzyme action of 30 peptides found in the PMF of the new protein. We implemented the present model in a server at portal Bio-AIMS (http://miaja.tic.udc.es/Bio-AIMS/EnzClassPred.php). This free on-line tool is based on PHP/HTML/Python and MARCH-INSIDE routines. This combined strategy may be used to identify and predict peptides of prokaryote and eukaryote parasites and their hosts as well as other superior organisms, which may be of interest in drug development or target identification.

Generalized lattice graphs for 2D-visualization of biological information

Abstract Several graph representations have been introduced for different data in theoretical biology. For instance, complex networks based on Graph theory are used to represent the structure and/or dynamics of different large biological systems such as protein–protein interaction networks. In addition, Randic, Liao, Nandy, Basak, and many others developed some special types of graph-based representations. This special type of graph includes geometrical constrains to node positioning in space and adopts final geometrical shapes that resemble lattice-like patterns. Lattice networks have been used to visually depict DNA and protein sequences but they are very flexible. However, despite the proved efficacy of new lattice-like graph/networks to represent diverse systems, most works focus on only one specific type of biological data. This work proposes a generalized type of lattice and illustrates how to use it in order to represent and compare biological data from different sources. We exemplify the following cases: protein sequence; mass spectra (MS) of protein peptide mass fingerprints (PMF); molecular dynamic trajectory (MDTs) from structural studies; mRNA microarray data; single nucleotide polymorphisms (SNPs); 1D or 2D-Electrophoresis study of protein polymorphisms and protein-research patent and/or copyright information. We used data available from public sources for some examples but for other, we used experimental results reported herein for the first time. This work may break new ground for the application of Graph theory in theoretical biology and other areas of biomedical sciences.

NL MIND-BEST : a web server for ligands and proteins discovery--theoretic-experimental study of proteins of Giardia lamblia and new compounds active against Plasmodium falciparum

There are many protein ligands and/or drugs described with very different affinity to a large number of target proteins or receptors. In this work, we selected Ligands or Drug-target pairs (DTPs/nDTPs) of drugs with high affinity/non-affinity for different targets. Quantitative Structure-Activity Relationships (QSAR) models become a very useful tool in this context to substantially reduce time and resources consuming experiments. Unfortunately most QSAR models predict activity against only one protein target and/or have not been implemented in the form of public web server freely accessible online to the scientific community. To solve this problem, we developed here a multi-target QSAR (mt-QSAR) classifier using the MARCH-INSIDE technique to calculate structural parameters of drug and target plus one Artificial Neuronal Network (ANN) to seek the model. The best ANN model found is a Multi-Layer Perceptron (MLP) with profile MLP 20:20-15-1:1. This MLP classifies correctly 611 out of 678 DTPs (sensitivity=90.12%) and 3083 out of 3408 nDTPs (specificity=90.46%), corresponding to training accuracy=90.41%. The validation of the model was carried out by means of external predicting series. The model classifies correctly 310 out of 338 DTPs (sensitivity=91.72%) and 1527 out of 1674 nDTP (specificity=91.22%) in validation series, corresponding to total accuracy=91.30% for validation series (predictability). This model favorably compares with other ANN models developed in this work and Machine Learning classifiers published before to address the same problem in different aspects. We implemented the present model at web portal Bio-AIMS in the form of an online server called: Non-Linear MARCH-INSIDE Nested Drug-Bank Exploration & Screening Tool (NL MIND-BEST), which is located at URL: http://miaja.tic.udc.es/Bio-AIMS/NL-MIND-BEST.php. This online tool is based on PHP/HTML/Python and MARCH-INSIDE routines. Finally we illustrated two practical uses of this server with two different experiments. In experiment 1, we report by first time Quantum QSAR study, synthesis, characterization, and experimental assay of antiplasmodial and cytotoxic activities of oxoisoaporphine alkaloids derivatives as well as NL MIND-BEST prediction of potential target proteins. In experiment 2, we report sampling, parasite culture, sample preparation, 2-DE, MALDI-TOF, and -TOF/TOF MS, MASCOT search, MM/MD 3D structure modeling, and NL MIND-BEST prediction for different peptides a new protein of the found in the proteome of the human parasite Giardia lamblia, which is promising for anti-parasite drug-targets discovery.

New Markov-Shannon Entropy models to assess connectivity quality in complex networks: from molecular to cellular pathway, Parasite-Host, Neural, Industry, and Legal-Social networks.

Graph and Complex Network theory is expanding its application to different levels of matter organization such as molecular, biological, technological, and social networks. A network is a set of items, usually called nodes, with connections between them, which are called links or edges. There are many different experimental and/or theoretical methods to assign node-node links depending on the type of network we want to construct. Unfortunately, the use of a method for experimental reevaluation of the entire network is very expensive in terms of time and resources; thus the development of cheaper theoretical methods is of major importance. In addition, different methods to link nodes in the same type of network are not totally accurate in such a way that they do not always coincide. In this sense, the development of computational methods useful to evaluate connectivity quality in complex networks (a posteriori of network assemble) is a goal of major interest. In this work, we report for the first time a new method to calculate numerical quality scores S(L(ij)) for network links L(ij) (connectivity) based on the Markov-Shannon Entropy indices of order k-th (θ(k)) for network nodes. The algorithm may be summarized as follows: (i) first, the θ(k)(j) values are calculated for all j-th nodes in a complex network already constructed; (ii) A Linear Discriminant Analysis (LDA) is used to seek a linear equation that discriminates connected or linked (L(ij)=1) pairs of nodes experimentally confirmed from non-linked ones (L(ij)=0); (iii) the new model is validated with external series of pairs of nodes; (iv) the equation obtained is used to re-evaluate the connectivity quality of the network, connecting/disconnecting nodes based on the quality scores calculated with the new connectivity function. This method was used to study different types of large networks. The linear models obtained produced the following results in terms of overall accuracy for network reconstruction: Metabolic networks (72.3%), Parasite-Host networks (93.3%), CoCoMac brain cortex co-activation network (89.6%), NW Spain fasciolosis spreading network (97.2%), Spanish financial law network (89.9%) and World trade network for Intelligent & Active Food Packaging (92.8%). In order to seek these models, we studied an average of 55,388 pairs of nodes in each model and a total of 332,326 pairs of nodes in all models. Finally, this method was used to solve a more complicated problem. A model was developed to score the connectivity quality in the Drug-Target network of US FDA approved drugs. In this last model the θ(k) values were calculated for three types of molecular networks representing different levels of organization: drug molecular graphs (atom-atom bonds), protein residue networks (amino acid interactions), and drug-target network (compound-protein binding). The overall accuracy of this model was 76.3%. This work opens a new door to the computational reevaluation of network connectivity quality (collation) for complex systems in molecular, biomedical, technological, and legal-social sciences as well as in world trade and industry.

Complex Network Spectral Moments for ATCUN Motif DNA Cleavage: First Predictive Study on Proteins of Human Pathogen Parasites

The development of methods that can predict the metal-mediated biological activity based only on the 3D structure of metal-unbound proteins has become a goal of major importance. This work is dedicated to the amino terminal Cu(II)- and Ni(II)-binding (ATCUN) motifs that participate in the DNA cleavage and have antitumor activity. We have calculated herein, for the first time, the 3D electrostatic spectral moments for 415 different proteins, including 133 potential ATCUN antitumor proteins. Using these parameters as input for Linear Discriminant Analysis, we have found a model that discriminates between ATCUN-DNA cleavage proteins and nonactive proteins with 91.32% Accuracy (379 out of 415 of proteins including both training and external validation series). Finally, the model has predicted for the first time the DNA cleavage function of proteins from the pathogen parasites. We have predicted possible ATCUN-like proteins with a probability higher than 99% in nine parasite families such as Trypanosoma, Plasmodium, Leishmania, or Toxoplasma. The distribution by biological function of the ATCUN proteins predicted has been the following: oxidoreductases 70.5%, signaling proteins 62.5%, lyases 58.2%, membrane proteins 45.5%, ligases 44.4%, hydrolases 41.3%, transferases 39.2%, cell adhesion proteins 34.5%, metal binders 33.5%, translation proteins 25.0%, transporters 16.7%, structural proteins 9.1%, and isomerases 8.2%. The model is implemented at http://miaja.tic.udc.es/Bio-AIMS/ATCUNPred.php.