Alisson Marques Silva

A fast learning algorithm for evolving neo-fuzzy neuron

This paper introduces an evolving neural fuzzy modeling approach constructed upon the neo-fuzzy neuron and network. The approach uses an incremental learning scheme to simultaneously granulate the input space and update the neural network weights. The neural network structure and parameters evolve simultaneously as data are input. Initially the space of each input variable is granulated using two complementary triangular membership functions. New triangular membership functions may be added, excluded and/or have their parameters adjusted depending on the input data and modeling error. The parameters of the network are updated using a gradient-based scheme with optimal learning rate. The performance of the approach is evaluated using instances of times series forecasting and nonlinear system identification problems. Computational experiments and comparisons against alternative evolving models show that the evolving neural neo-fuzzy network is accurate and fast, characteristics which are essential for adaptive systems modeling, especially in real-time, on-line environments.

Evolving Neo-fuzzy Neural Network with Adaptive Feature Selection

This paper suggests an approach to develop a class of evolving neural fuzzy networks with adaptive feature selection. The approach uses the neo-fuzzy neuron structure in conjunction with an incremental learning scheme that, simultaneously, selects the input variables, evolves the network structure, and updates the neural network weights. The mechanism of the adaptive feature selection uses statistical tests and information about the current model performance to decide if a new variable should be added, or if an existing variable should be excluded or kept as an input. The network structure evolves by adding or deleting membership functions and adapting its parameters depending of the input data and modeling error. The performance of the evolving neural fuzzy network with adaptive feature selection is evaluated considering instances of times series forecasting problems. Computational experiments and comparisons show that the proposed approach is competitive and achieves higher or as high performance as alternatives reported in the literature.

Successful application of virtual screening and molecular dynamics simulations against antimalarial molecular targets

The main challenge in the control of malaria has been the emergence of drug-resistant parasites. The presence of drug-resistant Plasmodium sp. has raised the need for new antimalarial drugs. Molecular modelling techniques have been used as tools to develop new drugs. In this study, we employed virtual screening of a pyrazol derivative (Tx001) against four malaria targets: plasmepsin-IV, plasmepsin-II, falcipain-II, and PfATP6. The receiver operating characteristic curves and area under the curve (AUC) were established for each molecular target. The AUC values obtained for plasmepsin-IV, plasmepsin-II, and falcipain-II were 0.64, 0.92, and 0.94, respectively. All docking simulations were carried out using AutoDock Vina software. The ligand Tx001 exhibited a better interaction with PfATP6 than with the reference compound (-12.2 versus -6.8 Kcal/mol). The Tx001-PfATP6 complex was submitted to molecular dynamics simulations in vacuum implemented on an NAMD program. The ligand Tx001 docked at the same binding site as thapsigargin, which is a natural inhibitor of PfATP6. Compound TX001 was evaluated in vitro with a P. falciparum strain (W2) and a human cell line (WI-26VA4). Tx001 was discovered to be active against P. falciparum (IC50 = 8.2 µM) and inactive against WI-26VA4 (IC50 > 200 µM). Further ligand optimisation cycles generated new prospects for docking and biological assays.

Real-time nonlinear modeling of a twin rotor MIMO system using evolving neuro-fuzzy network

This paper presents an evolving neuro-fuzzy network approach (eNFN) to model a twin rotor MIMO system (TRMS) with two degrees of freedom in real-time. The TRMS is a fast, nonlinear, open loop unstable time-varying dynamic system, with cross coupling between the rotors. Modeling and control of TRMS require high sampling rates, typically in the order of milliseconds. Actual laboratory implementation shows that eNFN is fast, effective, and accurately models the TRMS in real-time. The eNFN captures the TRMS system dynamics quickly, and develops precise low cost models from the point of view of time and space complexity. The results suggest eNFN as a potential candidate to model complex, fast time-varying dynamic systems in real-time.

Docking-based virtual screening of Brazilian natural compounds using the OOMT as the pharmacological target database

AbstractThe demand for new therapies has encouraged the development of faster and cheaper methods of drug design. Considering the number of potential biological targets for new drugs, the docking-based virtual screening (DBVS) approach has occupied a prominent role among modern strategies for identifying new bioactive substances. Some tools have been developed to validate docking methodologies and identify false positives, such as the receiver operating characteristic (ROC) curve. In this context, a database with 31 molecular targets called the Our Own Molecular Targets Data Bank (OOMT) was validated using the root-mean-square deviation (RMSD) and the area under the ROC curve (AUC) with two different docking methodologies: AutoDock Vina and DOCK 6. Sixteen molecular targets showed AUC values of >0.8, and those targets were selected for molecular docking studies. The drug-likeness properties were then determined for 473 Brazilian natural compounds that were obtained from the ZINC database. Ninety-six compounds showed similar drug-likeness property values to the marked drugs (positive values). These compounds were submitted to DBVS for 16 molecular targets. Our results showed that AutoDock Vina was more appropriate than DOCK 6 for performing DBVS experiments. Furthermore, this work suggests that three compounds—ZINC13513540, ZINC06041137, and ZINC1342926—are inhibitors of the three molecular targets 1AGW, 2ZOQ, and 3EYG, respectively, which are associated with cancer. Finally, since ZINC and the PDB were solely created to store biomolecule structures, their utilization requires the application of filters to improve the first steps of the drug development process. Graphical AbstractEvaluation of docking methods used for virtual screening

Evolving Neural Fuzzy Network with Adaptive Feature Selection

This paper introduces a neural fuzzy network approach for evolving system modeling. The approach uses neofuzzy neurons and a neural fuzzy structure monished with an incremental learning algorithm that includes adaptive feature selection. The feature selection mechanism starts considering one or more input variables from a given set of variables, and decides if a new variable should be added, or if an existing variable should be excluded or kept as an input. The decision process uses statistical tests and information about the current model performance. The incremental learning scheme simultaneously selects the input variables and updates the neural network weights. The weights are adjusted using a gradient-based scheme with optimal learning rate. The performance of the models obtained with the neural fuzzy modeling approach is evaluated considering weather temperature forecasting problems. Computational results show that the approach is competitive with alternatives reported in the literature, especially in on-line modeling situations where processing time and learning are critical.

MILKDE: A new approach for multiple instance learning based on positive instance selection and kernel density estimation

Abstract Multiple Instance Learning (MIL) is a recent paradigm of learning, which is based on the assignment of a single label to a set of instances called bag. A bag is positive if it contains at least one positive instance, and negative otherwise. This work proposes a new algorithm based on likelihood computation by means of Kernel Density Estimation (KDE) called MILKDE. Using the LogitBoost classifier, its performance was compared to that of forty-three MIL algorithms available in the literature using five data sets. Our proposal outperformed all of them for the Elephant (87.40%), Fox (66.80%) and COREL 2000 data sets (77.8%), and achieved competitive results for the MUSK 1 (89.20%) and MUSK 2 (87.50%) data sets, which are comparable to the higher accuracies obtained by other methods for this data sets. Overall results are statistically comparable to those obtained by the most well known methods for MIL described in the literature.

Artificial Neural Networks and Ranking Approach for Probe Selection and Classification of Microarray Data

Acute leukemia classification into its Myeloid and Lymphoblastic subtypes is usually accomplished according to the morphological appearance of the tumor. Nevertheless, cells from the two subtypes can have similar histopathological appearance, which makes screening procedures very difficult. Correct classification of patients in the initial phases of the disease would allow doctors to properly prescribe cancer treatment. Therefore, the development of alternative methods, to the usual morphological classification, is needed in order to improve classification rates and treatment. This paper is based on the principle that DNA microarray data extracted from tumors contain sufficient information to differentiate leukemia subtypes. The classification task is described as a general pattern recognition problem, requiring initial representation by causal quantitative features, followed by the construction of a classifier. In order to show the validity of our methods, a publicly available dataset of acute leukemia comprising 72 samples with 7,129 features was used. The dataset was split into two subsets: the training dataset with 38 samples and the test dataset with 34 samples. Feature selection methods were applied to the training dataset. The 50 most predictive genes, according to each method, were selected. Artificial Neural Network (ANN) classifiers were developed to compare the feature selection methods. Among the 50 genes selected using the best classifier, 21 are consistent with previous work and 4 additional ones are clearly related to tumor molecular processes. The remaining 25 selected genes were able to classify the test dataset, correctly, using the ANN.

Adaptive Input Selection and Evolving Neural Fuzzy Networks Modeling

AbstractThis paper suggests an evolving approach to develop neural fuzzy networks for system modeling. The approach uses an incremental learning procedure to simultaneously select the model inputs, to choose the neural network structure, and to update the network weights. Candidate models with larger and smaller number of input variables than the current model are constructed and tested concurrently. The procedure employs a statistical test in each learning step to choose the best model amongst the current and candidate models. Membership functions can be added or deleted to adjust input space granulation and the neural network structure. Granulation and structure adaptation depend of the modeling error. The weights of the neural networks are updated using a gradient-descent algorithm with optimal learning rate. Prediction and nonlinear system identification examples illustrate the usefulness of the approach. Comparisons with state of the art evolving fuzzy modeling alternatives are performed to evaluate ...