Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “Neural network”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,531 records · Page 85Linked to original sources
PubMed1995

Artificial neural network method for discriminating coding regions of eukaryotic genes.

This paper describes the application of artificial neural networks to discriminating the coding system of eukaryotic genes. We choose > 300 genes from eight eukaryotic organisms: human, mouse, rat, horse, ox, sheep, soybean and rabbit, from which we build up different discrimination models relevant to their promoter regions, poly(A) signals, splice site locations of introns and noose structures. The results shows that as long as the coding length is definite, the only correct coding region can be chosen from the large number of possible solutions discriminated by neural networks.

Animals↗
PubMed1999

Estimation of tumor stage and lymph node status in patients with colorectal adenocarcinoma using probabilistic neural networks and logistic regression.

Staging colorectal adenocarcinoma on the basis of biopsy specimens could identify patients who might benefit from neoadjuvant therapy without undergoing resection first. In this study, we evaluated the ability of artificial neural networks with genetic algorithms and multivariate logistic regression to predict the stage of 99 patients with primary colorectal adenocarcinoma by analyzing age, tumor grade, and immunoreactivity to p53 and bcl-2 with use of endoscopically obtained biopsy specimens. We correlated results with regional lymph node status and tumor stage, identified in subsequent colectomy specimens. bcl-2 and p53 protein expression were demonstrated by immunohistochemical methods, using formalin-fixed, paraffin-embedded biopsy tissues. Tumor grade was evaluated in hematoxylinand eosin-stained sections. Patients were divided into training (n = 75) and testing cases (n = 24). Several probabilistic neural networks with genetic algorithm models were trained, using the four prognostic features as input neurons and regional lymph node status or stage as output neurons. Data were analyzed with univariate statistics and multivariate logistic regression. The cases were divided into training (n = 40) and testing (n = 59). The best two models classified correctly the lymph node status of 20 of 24 test patients (specificity, 80%; sensitivity, 85%; positive predictive value, 86%) and the tumor stage of 21 of 24 test patients (specificity, 82%; sensitivity, 92%; positive predictive value, 85%), respectively. Tumor grade and p53 protein were statistically significant (P < .05) by analysis of variance for lymph node status and tumor stage. Logistic regression models with these two independent variables correctly estimated the probability of lymph node metastases in 44 of 59 test cases and the tumor stage of 43 of 59 test cases, respectively. Results indicated the usefulness of probabilistic neural networks in the population studied, but the findings should be validated with large groups of patients.

Adenocarcinoma↗
PubMed1998

Prediction of protein hydration sites from sequence by modular neural networks.

The hydration properties of a protein are important determinants of its structure and function. Here, modular neural networks are employed to predict ordered hydration sites using protein sequence information. First, secondary structure and solvent accessibility are predicted from sequence with two separate neural networks. These predictions are used as input together with protein sequences for networks predicting hydration of residues, backbone atoms and sidechains. These networks are trained with protein crystal structures. The prediction of hydration is improved by adding information on secondary structure and solvent accessibility and, using actual values of these properties, residue hydration can be predicted to 77% accuracy with a Matthews coefficient of 0.43. However, predicted property data with an accuracy of 60-70% result in less than half the improvement in predictive performance observed using the actual values. The inclusion of property information allows a smaller sequence window to be used in the networks to predict hydration. It has a greater impact on the accuracy of hydration site prediction for backbone atoms than for sidechains and for non-polar than polar residues. The networks provide insight into the mutual interdependencies between the location of ordered water sites and the structural and chemical characteristics of the protein residues.

Amino Acid Sequence↗
PubMed2006

Effects of fast presynaptic noise in attractor neural networks.

We study both analytically and numerically the effect of presynaptic noise on the transmission of information in attractor neural networks. The noise occurs on a very short timescale compared to that for the neuron dynamics and it produces short-time synaptic depression. This is inspired in recent neurobiological findings that show that synaptic strength may either increase or decrease on a short timescale depending on presynaptic activity. We thus describe a mechanism by which fast presynaptic noise enhances the neural network sensitivity to an external stimulus. The reason is that, in general, presynaptic noise induces nonequilibrium behavior and, consequently, the space of fixed points is qualitatively modified in such a way that the system can easily escape from the attractor. As a result, the model shows, in addition to pattern recognition, class identification and categorization, which may be relevant to the understanding of some of the brain complex tasks.

Animals↗
PubMed1998

The recursive deterministic perceptron neural network.

We introduce a feedforward multilayer neural network which is a generalization of the single layer perceptron topology (SLPT), called recursive deterministic perceptron (RDP). This new model is capable of solving any two-class classification problem, as opposed to the single layer perceptron which can only solve classification problems dealing with linearly separable sets (two subsets X and Y of R(d) are said to be linearly separable if there exists a hyperplane such that the elements of X and Y lie on the two opposite sides of R(d) delimited by this hyperplane). We propose several growing methods for constructing a RDP. These growing methods build a RDP by successively adding intermediate neurons (IN) to the topology (an IN corresponds to a SLPT). Thus, as a result, we obtain a multilayer perceptron topology, which together with the weights, are determined automatically by the constructing algorithms. Each IN augments the affine dimension of the set of input vectors. This augmentation is done by adding the output of each of these INs, as a new component, to every input vector. The construction of a new IN is made by selecting a subset from the set of augmented input vectors which is LS from the rest of this set. This process ends with LS classes in almost n-1 steps where n is the number of input vectors. For this construction, if we assume that the selected LS subsets are of maximum cardinality, the problem is proven to be NP-complete. We also introduce a generalization of the RDP model for classification of m classes (m>2) allowing to always separate m classes. This generalization is based on a new notion of linear separability for m classes, and it follows naturally from the RDP. This new model can be used to compute functions with a finite domain, and thus, to approximate continuous functions. We have also compared - over several classification problems - the percentage of test data correctly classified, or the topology of the 2 and m classes RDPs with that of the backpropagation (BP), cascade correlation (CC), and two other growing methods.

Journal Article↗
PubMed1999

Dynamics of neural networks with a central element.

A neural network is considered which is designed as a system of phase oscillators and contains a central oscillator that interacts with a number of peripheral oscillators. Analytical and simulation methods are used to study the dynamics of the system that is conditioned by the interaction parameters and natural frequencies of the oscillators. The boundaries of parameter regions are found that correspond to the synchronization of the whole network or to partial synchronization between the central oscillator and a group of peripheral oscillators. For a system with two peripheral oscillators the bifurcation analysis is applied to describe the changes of synchronization modes. The implications of the results for attention modeling are discussed.

Journal Article↗
PubMedDate not supplied

The neural network as a tool for multispectral interpretation.

A neural network which utilized data from the infrared spectra, carbon-13 NMR spectra, and molecular formulas of organic compounds was developed. The network, which had one layer of hidden units, was trained by backpropagation; network parameters were determined by a simplex optimization procedure. A database of 1560 compounds was used for training and testing. The trained network was able to identify with high accuracy the presence of a broad range of substructural features present in the compounds. The number of features identified and the accuracy were significantly greater as compared with networks using data from a single form of spectroscopy. The results have significance for the SESAMI computer-enhanced structure elucidation system.

Computers↗
PubMed2000

Nonlinear multivariate regression outperforms several concisely designed neural networks on three QSPR data sets

Neural networks (NNs) are accepted as the most powerful nonlinear technique in QSAR and QSPR modeling. However, the NN models are often very robust, containing a large number of parameters optimized during the training procedure. We have recently found (J. Chem. Inf. Comput. Sci. 1999, 39, 121-132) that the simpler nonlinear multiregression (MR) models are significantly better than the robust NNs, according to the same statistical parameters. In the present paper we investigated whether the nonlinear MR models are also better than the concisely designed NN models. Nonlinear MR models were generated in the following way. First, nonlinear terms, the 2-fold and 3-fold cross-products of initial descriptors, were calculated and added to initial descriptors. Then, the combination of two powerful techniques for descriptor selection (CROMRsel for "the best" selection and CROMRiisel for approximative, "i by i" stepwise selection) were used to detect the most important descriptors in MR models. For boiling points (BPs) of 150 alkanes the 20-descriptor MR model produced the cross-validated (CV) standard error of 2.88 K, and the best NN model (with 70-80 adjusted weights) had 3.60 K. Prediction of BPs of 50 compounds using the 17-descriptor MR model (obtained on 100 compounds) gave the standard error of 3.58 K. In the case of modeling of 243 chemical shifts CV standard errors were (in ppm) 0.89 and 1.19 with 15- and 9-descriptor MR models, respectively. The best NN models adjusted 60-90 weights and achieved 1.42 ppm. The standard error in predicting the 83 chemical shifts using the 10-descriptor MR model obtained on 160 samples was 1.25 ppm. It is also shown in this data set that the model quality depends on the scaling procedure used for transformation of the initial descriptors. In modeling the sublimation enthalpy the CV correlation coefficient was 0.97 using the best 4-descriptor MR model versus 0.93 obtained using NN with approximately 50 adjusted weights. The CV correlation coefficient in predicting the sublimation enthalpies for 21 compounds using the 4-descriptor MR model was 0.98. This is, to our knowledge, the first unambiguous result which shows a way for obtaining nonlinear MR models having better fitted, cross-validated, and predictive performances than the corresponding NN models. Moreover, the nonlinear MR models are significantly simpler than the NN models, which allows one to establish the functional relationships between the modeled property/activity and descriptors.

Journal Article↗
PubMed1990

Global processing of visual stimuli in a neural network of coupled oscillators.

An oscillator neural network model is presented that is capable of processing local and global attributes of sensory input. Local features in the input are encoded in the average firing rate of the neurons while the relationships between these features can modulate the temporal structure of the neuronal output. Neurons that share the same receptive field interact via relatively strong feedback connections, while neurons with different fields interact via specific, relatively weak connections. This pattern of connectivity mimics that of primary visual cortex. The model is studied in the context of processing visual stimuli that are coded for orientation. We compare our theoretical results with recent experimental evidence on coherent oscillatory activity in the cat visual cortex. The computational capabilities of the model for performing discrimination and segmentation tasks are demonstrated.

Animals↗
PubMed2002

Minimum description length neural networks for time series prediction.

Artificial neural networks (ANN) are typically composed of a large number of nonlinear functions (neurons) each with several linear and nonlinear parameters that are fitted to data through a computationally intensive training process. Longer training results in a closer fit to the data, but excessive training will lead to overfitting. We propose an alternative scheme that has previously been described for radial basis functions (RBF). We show that fundamental differences between ANN and RBF make application of this scheme to ANN nontrivial. Under this scheme, the training process is replaced by an optimal fitting routine, and overfitting is avoided by controlling the number of neurons in the network. We show that for time series modeling and prediction, this procedure leads to small models (few neurons) that mimic the underlying dynamics of the system well and do not overfit the data. We apply this algorithm to several computational and real systems including chaotic differential equations, the annual sunspot count, and experimental data obtained from a chaotic laser. Our experiments indicate that the structural differences between ANN and RBF make ANN particularly well suited to modeling chaotic time series data.

Journal Article↗
PubMedDate not supplied

Assessment of nodal involvement and survival analysis in breast cancer patients using image cytometric data: statistical, neural network and fuzzy approaches.

Accurate and reliable decision making in breast cancer prognosis can help in the planning of suitable surgery and therapy and, generally, optimise patient management through the different stages of the disease. In recent years, several prognostic factors have been used as indicators of disease progression in breast cancer. In this paper we investigate a fuzzy method, namely fuzzy k-nearest neighbour technique for breast cancer prognosis, and for determining the significance of prognostic markers and subsets of the markers, which include histology type, tumour grade, DNA ploidy, S-phase fraction, G0G1/G2M ratio, and minimum (start) and maximum (end) nuclear pleomorphism indices. We also compare the method with (a) logistic regression as a statistical method, and (b) multilayer feed forward backpropagation neural networks as an artificial neural network tool, the latter two techniques having been widely used for cancer prognosis. Nodal involvement and survival analyses in breast cancer are carried out for 100 women who were clinically diagnosed with breast disease in the form of carcinoma and benign conditions, and seven prognostic markers collected for each patient. For nodal involvement analysis, node positive and negative patients are predicted whereas survival analysis is carried out for two categories: whether a patient is alive or dead within 5 years of diagnosis. The results obtained show that the fuzzy method yields the highest predictive accuracy of 88% for both nodal involvement and survival analyses obtained from the subsets of [tumour grade, S-phase fraction, minimum (start) nuclear pleomorphism index] and [tumour histology type, DNA ploidy, S-phase fraction, G0G1/G2M ratio], respectively. We believe that this technique has produced more reliable prognostic factor models than those obtained using either the statistical or artificial neural networks-based methods.

Breast Neoplasms↗
PubMed2003

Lung function interpolation by means of neural-network-supported analysis of respiration sounds.

Respiration sounds of individual asthmatic patients were analysed in the scope of the development of a method for computerised recognition of the degree of airways obstruction. Respiration sounds were recorded during laboratory sessions of allergen provoked airways obstruction, during several stages of advancing obstruction. The technique of artificial neural networks was applied for relating sound spectra and simultaneously measured lung function values (spirometry parameter FEV(1)). The ability of feedforward neural networks was tested to interpolate obstruction levels of FEV(1)-classes of which no members were included in the set used for training a network. In this way, a situation was simulated of an existing network recognising a new asthmatic attack under the same physiological conditions. It appeared to be possible to interpolate FEV(1) values, and it is concluded that a deterministic relationship exists between sound spectra and lung function parameter FEV(1). Variance optimisation appeared to be important in optimising the neural network configuration.

Asthma↗
PubMed1991

Automated staging of sleep in cats using neural networks.

Manual staging of sleep based on visual EEG criteria is a laborious and time-consuming task. In an effort to automate sleep staging, we have developed a neural network that 'learns' to stage sleep on the basis of wave band count data alone, in the cat. Wave band count data are collected on a microcomputer, using period-amplitude analysis. Delta waves, spindle bursts, ponto-geniculo-occipital (PGO) waves, electro-oculogram (EOG), basal electromyogram (EMG) amplitude, and movement artifact amplitude are collected, and used to 'train' the network to score sleep. These wave count data serve as the input patterns to the net, and the corresponding manually scored sleep stages serve as a 'teacher.' We demonstrate that, when used to score the states of wake, slow wave sleep (SWS), desynchronized sleep (D), and the transition period from SWS to D (SP), these neural networks agree with manual scoring an average of 93.3% for all epochs scored. Neural network programs can learn both rules and exceptions, and since the nets teach themselves these rules automatically, a minimum of human effort is required. Because programming requirements are small for neural nets, this approach is readily adaptable to microcomputer-based systems and is widely applicable to both animal and human EEG analyses. The utility of this approach for the detection and classification of a variety of clinical neurophysiological disorders is discussed.

Animals↗
PubMed2006

A nested molecule-independent neural network approach for high-quality potential fits.

It is shown that neural networks (NNs) are efficient and effective tools for fitting potential energy surfaces. For H2O, a simple NN approach works very well. To fit surfaces for HOOH and H2CO, we develop a nested neural network technique in which we first fit an approximate NN potential and then use another NN to fit the difference of the true potential and the approximate potential. The root-mean-square error (RMSE) of the H2O surface is 1 cm(-1). For the 6-D HOOH and H2CO surfaces, the nested approach does almost as well attaining a RMSE of 2 cm(-1). The quality of the NN surfaces is verified by calculating vibrational spectra. For all three molecules, most of the low-lying levels are within 1 cm(-1) of the exact results. On the basis of these results, we propose that the nested NN approach be considered a method of choice for both simple potentials, for which it is relatively easy to guess a good fitting function, and complicated (e.g., double well) potentials for which it is much harder to deduce an appropriate fitting function. The number of fitting parameters is only moderately larger for the 6-D than for the 3-D potentials, and for all three molecules, decreasing the desired RMSE increases only slightly the number of required fitting parameters (nodes). NN methods, and in particular the nested approach we propose, should be good universal potential fitting tools.

Journal Article↗
PubMed2006

On the consistency of the blocked neural network estimator in time series analysis.

We describe a nonlinear regression problem, where the regression functions have an additive structure and the dependent variable is a one-dimensional time series. Multivariate time series with unknown time delay operators are used as independent variables. By fitting a feedforward neural network with block structure to the data, we estimated the additive regression function and, parallel to this, the time lags. We present the consistency proof of neural network weights estimator and the time lag estimator independently from each other. In the practical part of the article, we present the useful feature of blocked neural networks to estimate the relevance measures of each input variable in a simple way. Furthermore, we propose an approach to solve the well-known variable selection problem for the class of nonlinear multivariate beta-mixing time series models considered here. Finally, we apply the methodology to an artificial example.

Animals↗
PubMed2005

A novel wavelet neural network based pathological stage detection technique for an oral precancerous condition.

AIM: To describe a novel neural network based oral precancer (oral submucous fibrosis; OSF) stage detection method. METHOD: The wavelet coefficients of transmission electron microscopy images of collagen fibres from normal oral submucosa and OSF tissues were used to choose the feature vector which, in turn, was used to train the artificial neural network. RESULTS: The trained network was able to classify normal and oral precancer stages (less advanced and advanced) after obtaining the image as an input. CONCLUSIONS: The results obtained from this proposed technique were promising and suggest that with further optimisation this method could be used to detect and stage OSF, and could be adapted for other conditions.

Algorithms↗
PubMed2002

A neural network model of the inferior colliculus with modifiable lateral inhibitory synapses for human echolocation.

We propose a neural network model of the inferior colliculus (IC) for human echolocation. Neuronal mechanisms for human echolocation were investigated by simulating the model. The model consists of the neural networks of the central nucleus (ICc) and external nucleus (ICx) of the inferior colliculus. The neurons of the ICc receive interaural sound stimuli via multiple contralateral delay lines and a single ipsilateral delay line. The neurons of the ICc send output signals to the neurons of the ICx in a convergent manner. We stimulated the ICc with pairs of a direct sound (a sonar sound) and an echo sound (the reflection from an object). Information about the distance between the model and the object is expressed by the delay time of the echo sound with respect to the direct sound. The results presented here show that neurons of the ICc responsive to interaural onset time differences contribute to the creation of an auditory distance map in the ICx. We trained the model with various pairs of direct-echo sounds and modified synaptic connection strengths of the networks according to the Hebbian rule. It is shown that self-organized long-term depression of lateral inhibitory synaptic connections plays an important role in enhancing echolocation skills.

Acoustic Stimulation↗
PubMed2003

Entropy learning and relevance criteria for neural network pruning.

In this paper, entropy is a term used in the learning phase of a neural network. As learning progresses, more hidden nodes get into saturation. The early creation of such hidden nodes may impair generalisation. Hence an entropy approach is proposed to dampen the early creation of such nodes by using a new computation called entropy cycle. Entropy learning also helps to increase the importance of relevant nodes while dampening the less important nodes. At the end of learning, the less important nodes can then be pruned to reduce the memory requirements of the neural network.

Algorithms↗