PubMed Health⌕ Search

SEARCH · PubMed Health

Results for “Neural Networks, Computer”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

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

At least 163 records · Page 9Linked to original sources
PubMed1991

Generalized protein tertiary structure recognition using associative memory Hamiltonians.

In previous papers, a method of protein tertiary structure recognition was described based on the construction of an associative memory Hamiltonian, which encoded the amino acid sequence and the C alpha co-ordinates of a set of database proteins. Using molecular dynamics with simulated annealing, the ability of the Hamiltonian to successfully recall the structure of a protein in the memory database was successfully demonstrated, as long as the total number of database proteins did not exceed a characteristic value, called the capacity of the Hamiltonian, equal to 0.5N to 0.7N, where N is the number of amino acid residues in the protein to be recalled. In this paper, we describe the development of additional methods to increase the capacity of the Hamiltonian, including use of a more complete representation of the protein backbone and the incorporation of contextual information into the Hamiltonian through the use of secondary structure prediction. In addition, we further extend the ability of associative memory models to predict the tertiary structures of proteins not present in the protein data set, by making the Hamiltonian invariant with respect to biological symmetries that represent site mutations and insertions and deletions. The ability of the Hamiltonian to generalize from homologous proteins to an unknown protein in the presence of other unrelated proteins in the data set is demonstrated.

Cytochromes↗
PubMed1992

G+C-rich tract in 5' end of human introns.

Analysis of an artificial neural network trained to classify DNA as coding or non-coding revealed compositional differences between sequence parts translated into protein and those that were not. The 5' end of human introns was found to have a base composition that was non-random to an extent matching the non-randomness in the 3' end that contains the polypyrimidine tract. The prevailing nucleotides in the initial 50 nucleotides of human introns are guanine and cytosine, the trinucleotide GGG was found to occur almost four times as frequently as it would in sequences with a uniform distribution of the nucleotides. The initial part of terminal exons and their associated terminal introns were shown to have a very special base composition deviating strongly from the normal picture in other exons and introns.

Base Composition↗
PubMed1992

Predicting protein secondary structure using neural net and statistical methods.

A comparison of neural network methods and Bayesian statistical methods is presented for prediction of the secondary structure of proteins given their primary sequence. The Bayesian method makes the unphysical assumption that the probability of an amino acid occurring in each position in the protein is independent of the amino acids occurring elsewhere. However, we find the predictive accuracy of the Bayesian method to be only minimally less than the accuracy of the most sophisticated methods used to date. We present the relationship of neural network methods to Bayesian statistical methods and show that, in principle, neural methods offer considerable power, although apparently they are not particularly useful for this problem. In the process, we derive a neural formalism in which the output neurons directly represent the conditional probabilities of structure class. The probabilistic formalism allows introduction of a new objective function, the mutual information, which translates the notion of correlation as a measure of predictive accuracy into a useful training measure. Although a similar accuracy to other approaches (utilizing a mean-square error) is achieved using this new measure, the accuracy on the training set is significantly and tantalizingly higher, even though the number of adjustable parameters remains the same. The mutual information measure predicts a greater fraction of helix and sheet structures correctly than the mean-square error measure, at the expense of coil accuracy, precisely as it was designed to do. By combining the two objective functions, we obtain a marginally improved accuracy of 64.4%, with Matthews coefficients C alpha, C beta and Ccoil of 0.40, 0.32 and 0.42, respectively. However, since all methods to date perform only slightly better than the Bayes algorithm, which entails the drastic assumption of independence of amino acids, one is forced to conclude that little progress has been made on this problem, despite the application of a variety of sophisticated algorithms such as neural networks, and that further advances will require a better understanding of the relevant biophysics.

Bayes Theorem↗
PubMed1991

The principle of interval constraints: a generalization of the symmetric Dirichlet distribution.

A structure for representing problems in decision analysis and in expert systems, which reason under uncertainty, is the influence diagram or causal network. A causal network consists of an underlying joint probability distribution and a directed acyclic graph in which a propositional variable that represents a marginal distribution is stored at each vertex in the graph. This paper is concerned with two of the problems in applications that use causal networks. The first problem is the determination of the conditional probabilities of the values of remaining propositional variables in the network given that certain variables are instantiated for particular values. This is called probability propagation. The second problem is the determination of the most probable, second most probable, third most probable, and so on sets of values of a particular set of variables (called the explanation set) given that certain variables are instantiated for particular values. This problem is called abductive inference. There exists a class of causal networks in which each variable has only two parents, for which the time is required, by any known method, for probability propagation is exponential relative to the number of vertices in the network. The determination of a new method that would be efficient for all causal networks appears unlikely, because probability propagation has been shown to be #P-complete. In many medical applications, networks are often large and not sparsely connected. Therefore a method for the exact determination of probability values appears unlikely for such applications, and the development of approximation methods seems to be the best solution. The current approximation methods obtain interval bounds for the probability values. When such intervals are obtained, it is not possible in general to rank the alternatives. In this paper, a method is developed for obtaining expected values for the point probabilities from interval constraints on the probabilities. The method is based on an application of the principle of indifference to the probability values themselves. The distributions obtained with the principle of indifference are a generalization of the symmetric Dirichlet distribution in which prior ignorance is assumed.

Decision Making↗
PubMed1991

Ecosystem flow networks: loaded dice?

An information-theoretic comparison of the topologies of observed ecosystem transfers and randomly constructed networks reveals that it is not easy to separate the members of the two sets. The distribution of ecosystem flow magnitudes, however, is seen to differ markedly from ordinary probability functions and to resemble the Cauchy or Pareto distributions. The agencies that impart such structure to ecological flow networks are not obvious, but one strong possibility is that autocatalysis, or indirect mutualism, promotes certain pathways at the expense of others, thereby enlarging the tail of the distribution of flow magnitudes.

Ecology↗
PubMed1992

Do biomolecules process information differently than synthetic organic molecules?

This paper compares information/signal processing in synthetic and biological molecules. The role of conformation-based (shape-based) mechanisms and electrostatic interactions in molecular recognition is discussed. In biological electron transfer, the 'electron shuttle'-mediated mechanism is contrasted with the mechanism based on pre-formed 'electron wires'. While biological information processing is thought to be more distributed (less discrete), an example of molecular switch is presented: visual transduction. We further speculate that visual transduction may be implemented in the form of a switch based on electrostatic interactions. The concept of intelligent materials is discussed with the well-known Bohr effect of hemoglobin oxygenation. Based on these examples, we argue that there are no fundamental differences between synthetic and biological molecules in their mode of information processing. In the pursuit of novel paradigms of molecular information processing, we also perceive no conflicts in developing molecular devices that emulate the switching function of conventional microelectronic devices.

Artificial Intelligence↗
PubMed1992

Molecules as electronic components?

Aspects of electronics at the molecular level are reviewed. Molecules can store information in their different states. To transmit information, they need to interact with other molecules, but this affects the states. Excitation transfer is also more complex than sometimes realised. Information processing is best treated in terms of the whole system of interacting molecules, so that molecules cannot be treated simply as small conventional electronic components.

Electrochemistry↗
PubMed1991

A neural network model of neural activity in the monkey globus pallidus.

A 3-layer neural network model was constructed to determine the discharge patterns of neurons within the globus pallidus (GP) which would be required to run a sequence of movements at the motor cortical level. The model was based on the presence of tonic and phasic neuronal activity within the motor cortical region; that positive feedback was required to maintain tonic neuronal activity and that phasic neuronal activity was required to initiate and terminate the tonic neuronal activity. The model predicted the presence of both phasic and tonic activity within the middle layer (layer 2; GP) of the model in order for the motor cortical regions (layers 1 and 3) to be able to run and to maintain the movement sequence. This prediction was in keeping with our electrophysiological findings within GP.

Animals↗
PubMed1991

On the possible physical foundations of health related effects of crystals.

I propose a possible physical mechanism to account for widely claimed health related effects of crystals. The hypothesis relates the mechanism of such action to the isotropic diversity of crystals which leads to an enormous information storage capacity of crystal lattices. Sublattices of particular isotopes form an interactive network similar to simulated neural networks based on spin glass models.

Crystallization↗
PubMed1988

The role of the teacher in learning-based models of parietal area 7a.

The back-propagation learning procedure can be used to train simulated neural networks to compute arbitrary functions. We have recently shown that when such a network is trained to carry out the transformation of stimulus location to head-centered coordinates that occurs in parietal area 7a, the response properties of certain units in the network closely resemble neurons found in area 7a. The back-propagation procedure requires the use of a teacher. Here we examine the effect of using different kinds of teachers. As long as the teacher represents information about stimulus location in head-centered coordinates, the trained network contains units of the kind found in area 7a. Differences in teacher format only effect the quantitative distribution of the different unit types. When the teacher does not represent stimulus location explicitly, the network does not contain units of the required kind.

Computer Simulation↗
PubMed2025

Comparative analysis of convolutional neural network models for the histopathological differentiation of acinic cell carcinoma and secretory carcinoma.

OBJECTIVE: Although artificial intelligence tools show promise for enhancing the diagnosis of head and neck lesions, few studies have tested these resources for the microscopic diagnosis of salivary gland tumors. Specifically, the microscopic differentiation between acinic cell carcinoma and secretory carcinoma has never been addressed in this context. Therefore, this exploratory study aimed to comparatively evaluate the feasibility of applying convolutional neural networks for the microscopic differentiation between acinic cell carcinomas and secretory carcinomas. METHODS: A cross-sectional study using whole-slide images from 46 patients with acinic cell carcinoma (n = 26) or secretory carcinoma (n = 20) was conducted. Eight CNNs (ResNet-50, InceptionV3, VGG16, Xception, MobileNet, DenseNet121, EfficientNetB0, and EfficientNetV2B0) were trained and evaluated for accuracy, sensitivity, specificity, F1-score, and AUC. Performance was measured in training, validation, and test subsets. Accuracy and loss curves were also presented. RESULTS: InceptionV3 demonstrated the best overall performance, with the lowest loss (1.39), highest accuracy (0.81), sensitivity (0.90), and F1-score (0.81). VGG16 achieved the highest AUC (0.86) and precision (0.77). DenseNet121 showed the lowest performance in terms of accuracy (0.65) and F1-Score (0.52), but the highest specificity (0.85). CONCLUSION: This proof-of-concept study suggests that convolutional neural networks may be feasible tools to support the microscopic differentiation between acinic cell carcinoma and secretory carcinoma. The performance of these models critically depends on the size of the dataset and the quality of annotations. The findings should be interpreted cautiously given the limited dataset and potential sources of bias. Further validation with larger, multicenter datasets is needed before any clinical application can be considered.

Humans↗
PubMed2025

A multi-modal transformer for cell type-agnostic regulatory predictions.

Sequence-based deep learning models have emerged as powerful tools for deciphering the cis-regulatory grammar of the human genome but cannot generalize to unobserved cellular contexts. Here, we present EpiBERT, a multi-modal transformer that learns generalizable representations of genomic sequence and cell type-specific chromatin accessibility through a masked accessibility-based pre-training objective. Following pre-training, EpiBERT can be fine-tuned for gene expression prediction, achieving accuracy comparable to the sequence-only Enformer model, while also being able to generalize to unobserved cell states. The learned representations are interpretable and useful for predicting chromatin accessibility quantitative trait loci (caQTLs), regulatory motifs, and enhancer-gene links. Our work represents a step toward improving the generalization of sequence-based deep neural networks in regulatory genomics.

Humans↗
PubMed1991

A connectionist model of development.

We present a phenomenological modeling framework for development. Our purpose is to provide a systematic method for discovering and expressing correlations in experimental data on gene expression and other developmental processes. The modeling framework is based on a connectionist or "neural net" dynamics for biochemical regulators, coupled to "grammatical rules" which describe certain features of the birth, growth, and death of cells, synapses and other biological entities. We outline how spatial geometry can be included, although this part of the model is not complete. As an example of the application of our results to a specific biological system, we show in detail how to derive a rigorously testable model of the network of segmentation genes operating in the blastoderm of Drosophila. To further illustrate our methods, we sketch how they could be applied to two other important developmental processes: cell cycle control and cell-cell induction. We also present a simple biochemical model leading to our assumed connectionist dynamics which shows that the dynamics used is at least compatible with known chemical mechanisms.

Animals↗
PubMed1992

Analysis of the clinical variables driving decision in an artificial neural network trained to identify the presence of myocardial infarction.

STUDY OBJECTIVE: To determine which clinical variables drive the output of an artificial neural network trained to identify the presence of myocardial infarction. DESIGN: Partial output analysis. SETTING: Tertiary university teaching center. PARTICIPANTS: Seven hundred six patients more than 18 years old presenting with anterior chest pain. MEASUREMENTS: Differential network output analysis. MAIN RESULTS: A methodology was developed as the first step in measuring the impact input clinical variables have on the output (diagnosis) of an artificial neural network trained to identify the presence of acute myocardial infarction. The methodology revealed that the network used the presence of ECG findings, as well as the presence of rales, syncope, jugular venous distension, response to trinitroglycerin, and nausea and vomiting, as major predictive sources. Although this first-step analysis studied individual variables, it must be stated that the network comes to clinical closure based on the settings of all variables in a pattern and that the impact of a single variable cannot be taken out of the context of a pattern. CONCLUSION: An artificial neural network trained to recognize the presence of myocardial infarction appears to place diagnostic importance on clinical variables that have not been shown previously to be highly predictive for infarction.

Adolescent↗
PubMed1992

Prediction of structural and functional features of protein and nucleic acid sequences by artificial neural networks.

The applications of artificial neural networks to the prediction of structural and functional features of protein and nucleic acid sequences are reviewed. A brief introduction to neural networks is given, including a discussion of learning algorithms and sequence encoding. The protein applications mostly involve the prediction of secondary and tertiary structure from sequence. The problems in nucleic acid analysis tackled by neural networks are the prediction of translation initiation sites in Escherichia coli, the recognition of splice junctions in human mRNA, and the prediction of promoter sites in E. coli. The performance of the approach is compared with other current statistical methods.

Algorithms↗