Computational neural networks: Enhancing supervised learning algorithms via self-organization

A neural network processing scheme is proposed which utilizes a self-organizing Kohonen feature map as the front end to a feedforward classifier network. The results of a series of benchmarking studies based upon artificial statistical pattern recognition tasks indicate that the proposed architecture performs significantly better than conventional feedforward classifier networks when the decision regions are disjoint. This is attributed to the fact that the self-organization process allows internal units in the succeeding classifier network to be sensitive to a specific set of features in the input space at the outset of training.     

Topology representing network enables highly accurate classification of protein images taken by cryo electron-microscope without masking

In single-particle analysis, a three-dimensional (3-D) structure of a protein is constructed using electron microscopy (EM). As these images are very noisy in general, the primary process of this 3-D reconstruction is the classification of images according to their Euler angles, the images in each classified group then being averaged to reduce the noise level. In our newly developed strategy of classification, we introduce a topology representing network (TRN) method. It is a modified method of a growing neural gas network (GNG). In this system, a network structure is automatically determined in response to the images input through a growing process. After learning without a masking procedure, the GNG creates clear averages of the inputs as unit coordinates in multi-dimensional space, which are then utilized for classification. In the process, connections are automatically created between highly related units and their positions are shifted where the inputs are distributed in multi-dimensional space. Consequently, several separated groups of connected units are formed. Although the interrelationship of units in this space are not easily understood, we succeeded in solving this problem by converting the unit positions into two-dimensional (2-D) space, and by further optimizing the unit positions with the simulated annealing (SA) method. In the optimized 2-D map, visualization of the connections of units provided rich information about clustering. As demonstrated here, this method is clearly superior to both the multi-variate statistical analysis (MSA) and the self-organizing map (SOM) as a classification method and provides a first reliable classification method which can be used without masking for very noisy images.     

Self-organization of neurons described by the maximum-entropy principle

In the article the maximum-entropy principle and Parzen windows are applied to derive an optimal mapping of a continuous into a descrete random variable. The mapping can be performed by a network of self-organizing information processing units similar to biological neurons. Each neuron is selectively sensitized to one prototype from the sample space of the discrete random variable. The continuous random variable is applied as the input signal exciting the neurons. The response of the network is described by the excitation vector which represents the encoded input signal. Due to the interaction between neurons adaptive changes of prototypes are caused by the excitations. The derived mathematical model explains this interaction in detail; a simplified self-organization rule derived from it corresponds to that of Kohonen. One and two-dimensional examples of self-organization simulated on a computer are shown in the article.     

A Cellular Automata-Based Mathematical Model for Thymocyte Development

Intrathymic T cell development is an important process necessary for the normal formation of cell-mediated immune responses. Importantly, such a process depends on interactions of developing thymocytes with cellular and extracellular elements of the thymic microenvironment. Additionally, it includes a series of oriented and tunely regulated migration events, ultimately allowing mature cells to cross endothelial barriers and leave the organ. Herein we built a cellular automata-based mathematical model for thymocyte migration and development. The rules comprised in this model take into account the main stages of thymocyte development, two-dimensional sections of the normal thymic microenvironmental network, as well as the chemokines involved in intrathymic cell migration. Parameters of our computer simulations with further adjusted to results derived from previous experimental data using sub-lethally irradiated mice, in which thymus recovery can be evaluated. The model fitted with the increasing numbers of each CD4/CD8-defined thymocyte subset. It was further validated since it fitted with the times of permanence experimentally ascertained in each CD4/CD8-defined differentiation stage. Importantly, correlations using the whole mean volume of young normal adult mice revealed that the numbers of cells generated in silico with the mathematical model fall within the range of total thymocyte numbers seen in these animals. Furthermore, simulations made with a human thymic epithelial network using the same mathematical model generated similar profiles for temporal evolution of thymocyte developmental stages. Lastly, we provided in silico evidence that the thymus architecture is important in the thymocyte development, since changes in the epithelial network result in different theoretical profiles for T cell development/migration. This model likely can be used to predict thymocyte evolution following therapeutic strategies designed for recovery of the thymus in diseases coursing with thymus involution, such as some primary immunodeficiencies, acute infections, and malnutrition.     

Astrocyte- neuron interaction as a mechanism responsible for generation of neural synchrony: a study based on modeling and experiments

Neural synchronization is considered as an important mechanism for information processing. In addition, based on recent neurophysiologic findings, it is believed that astrocytes regulate the synaptic transmission of neuronal networks. Therefore, the present study focused on determining the functional contribution of astrocytes in neuronal synchrony using both computer simulations and extracellular field potential recordings. For computer simulations, as a first step, a minimal network model is constructed by connecting two Morris-Lecar neuronal models. In this minimal model, astrocyte-neuron interactions are considered in a functional-based procedure. Next, the minimal network is extended and a biologically plausible neuronal population model is developed which considers functional outcome of astrocyte-neuron interactions too. The employed structure is based on the physiological and anatomical network properties of the hippocampal CA1 area. Utilizing these two different levels of modeling, it is demonstrated that astrocytes are able to change the threshold value of transition from synchronous to asynchronous behavior among neurons. In this way, variations in the interaction between astrocytes and neurons lead to the emergence of synchronous/asynchronous patterns in neural responses. Furthermore, population spikes are recorded from CA1 pyramidal neurons in rat hippocampal slices to validate the modeling results. It demonstrates that astrocytes play a primary role in neuronal firing synchronicity and synaptic coordination. These results may offer a new insight into understanding the mechanism by which astrocytes contribute to stabilizing neural activities.     

An evolving neural network for the interpretation of gene expression patterns

It is of central interest in biology to understand how gene activity networks are coordinated and integrated in the cell. Within the field of genomics, microarray technologies have become a powerful technique for monitoring simultaneously the expression patterns of thousands of genes under different sets of conditions. A main task now is to propose analytical methods that can suggest which groups of genes are activated by similar conditions. We review several techniques based on self-organizing map and clustering algorithms but implemented through a network of units controlled by biologically inspired functions (see Table 1). The computer tool, named NBIA, permits a categorization that generates a set of gene groups with coordinated expression patterns.     

The competition between protein folding and aggregation: off-lattice minimalist model studies

Protein aggregation has been associated with a number of human diseases, and is a serious problem in the manufacture of recombinant proteins. Of particular interest to the biotechnology industry is deleterious aggregation that occurs during the refolding of proteins from inclusion bodies. As a complement to experimental efforts, computer simulations of multi-chain systems have emerged as a powerful tool to investigate the competition between folding and aggregation. Here we report results from Langevin dynamics simulations of minimalist model proteins. Order parameters are developed to follow both folding and aggregation. By mapping natural units to real units, the simulations are shown to be carried out under experimentally relevant conditions. Data pertaining to the contacts formed during the association process show that multiple mechanisms for aggregation exist, but certain pathways are statistically preferred. Kinetic data show that there are multiple time scales for aggregation, although most association events take place at times much shorter than those required for folding. Last, we discuss results presented here as a basis for future work aimed at rational design of mutations to reduce aggregation propensity, as well as for development of small-molecular weight refolding enhancers.     

A Self-Organizing Model of the Visual Development of Hand-Centred Representations

We show how hand-centred visual representations could develop in the primate posterior parietal and premotor cortices during visually guided learning in a self-organizing neural network model. The model incorporates trace learning in the feed-forward synaptic connections between successive neuronal layers. Trace learning encourages neurons to learn to respond to input images that tend to occur close together in time. We assume that sequences of eye movements are performed around individual scenes containing a fixed hand-object configuration. Trace learning will then encourage individual cells to learn to respond to particular hand-object configurations across different retinal locations. The plausibility of this hypothesis is demonstrated in computer simulations.     

Optimization of an Elastic Network Augmented Coarse Grained Model to Study CCMV Capsid Deformation

The major protective coat of most viruses is a highly symmetric protein capsid that forms spontaneously from many copies of identical proteins. Structural and mechanical properties of such capsids, as well as their self-assembly process, have been studied experimentally and theoretically, including modeling efforts by computer simulations on various scales. Atomistic models include specific details of local protein binding but are limited in system size and accessible time, while coarse grained (CG) models do get access to longer time and length scales but often lack the specific local interactions. Multi-scale models aim at bridging this gap by systematically connecting different levels of resolution. Here, a CG model for CCMV (Cowpea Chlorotic Mottle Virus), a virus with an icosahedral shell of 180 identical protein monomers, is developed, where parameters are derived from atomistic simulations of capsid protein dimers in aqueous solution. In particular, a new method is introduced to combine the MARTINI CG model with a supportive elastic network based on structural fluctuations of individual monomers. In the parametrization process, both network connectivity and strength are optimized. This elastic-network optimized CG model, which solely relies on atomistic data of small units (dimers), is able to correctly predict inter-protein conformational flexibility and properties of larger capsid fragments of 20 and more subunits. Furthermore, it is shown that this CG model reproduces experimental (Atomic Force Microscopy) indentation measurements of the entire viral capsid. Thus it is shown that one obvious goal for hierarchical modeling, namely predicting mechanical properties of larger protein complexes from models that are carefully parametrized on elastic properties of smaller units, is achievable.     

Geometrically repatterned immunological synapses uncover formation mechanisms

The interaction of T cells and antigen-presenting cells is central to adaptive immunity and involves the formation of immunological synapses in many cases. The surface molecules of the cells form a characteristic spatial pattern whose formation mechanisms and function are largely unknown. We perform computer simulations of recent experiments on geometrically repatterned immunological synapses and explain the emerging structure as well as the formation dynamics. Only the combination of in vitro experiments and computer simulations has the potential to pinpoint the kind of interactions involved. The presented simulations make clear predictions for the structure of the immunological synapse and elucidate the role of a self-organizing attraction between complexes of T cell receptor and peptide–MHC molecule, versus a centrally directed motion of these complexes.     

Timing in the absence of clocks: encoding time in neural network states

Decisions based on the timing of sensory events are fundamental to sensory processing. However, the mechanisms by which the brain measures time over ranges of milliseconds to seconds remain unclear. The dominant model of temporal processing proposes that an oscillator emits events that are integrated to provide a linear metric of time. We examine an alternate model in which cortical networks are inherently able to tell time as a result of time-dependent changes in network state. Using computer simulations we show that within this framework, there is no linear metric of time, and that a given interval is encoded in the context of preceding events. Human psychophysical studies were used to examine the predictions of the model. Our results provide theoretical and experimental evidence that, for short intervals, there is no linear metric of time, and that time may be encoded in the high-dimensional state of local neural networks.     

A natural coarse graining for simulating large biomolecular motion

Various coarse graining schemes have been proposed to speed up computer simulations of the motion within large biomolecules, which can contain hundreds of thousands of atoms. We point out here that there is a very natural way of doing this, using the rigid regions identified within a biomolecule as the coarse grain elements. Subsequently, computer resources can be concentrated on the flexible connections between the rigid units. Examples of the use of such techniques are given for the protein barnase and the maltodextrin binding protein, using the geometric simulation technique FRODA and the rigidity enhanced elastic network model RCNMA to compute mobilities and atomic displacements.     

A learning theory for reward-modulated spike-timing-dependent plasticity with application to biofeedback

Reward-modulated spike-timing-dependent plasticity (STDP) has recently emerged as a candidate for a learning rule that could explain how behaviorally relevant adaptive changes in complex networks of spiking neurons could be achieved in a self-organizing manner through local synaptic plasticity. However, the capabilities and limitations of this learning rule could so far only be tested through computer simulations. This article provides tools for an analytic treatment of reward-modulated STDP, which allows us to predict under which conditions reward-modulated STDP will achieve a desired learning effect. These analytical results imply that neurons can learn through reward-modulated STDP to classify not only spatial but also temporal firing patterns of presynaptic neurons. They also can learn to respond to specific presynaptic firing patterns with particular spike patterns. Finally, the resulting learning theory predicts that even difficult credit-assignment problems, where it is very hard to tell which synaptic weights should be modified in order to increase the global reward for the system, can be solved in a self-organizing manner through reward-modulated STDP. This yields an explanation for a fundamental experimental result on biofeedback in monkeys by Fetz and Baker. In this experiment monkeys were rewarded for increasing the firing rate of a particular neuron in the cortex and were able to solve this extremely difficult credit assignment problem. Our model for this experiment relies on a combination of reward-modulated STDP with variable spontaneous firing activity. Hence it also provides a possible functional explanation for trial-to-trial variability, which is characteristic for cortical networks of neurons but has no analogue in currently existing artificial computing systems. In addition our model demonstrates that reward-modulated STDP can be applied to all synapses in a large recurrent neural network without endangering the stability of the network dynamics.     

Development of neural circuitry for precise temporal sequences through spontaneous activity, axon remodeling, and synaptic plasticity

Temporally precise sequences of neuronal spikes that span hundreds of milliseconds are observed in many brain areas, including songbird premotor nucleus, cat visual cortex, and primary motor cortex. Synfire chains-networks in which groups of neurons are connected via excitatory synapses into a unidirectional chain-are thought to underlie the generation of such sequences. It is unknown, however, how synfire chains can form in local neural circuits, especially for long chains. Here, we show through computer simulation that long synfire chains can develop through spike-time dependent synaptic plasticity and axon remodeling-the pruning of prolific weak connections that follows the emergence of a finite number of strong connections. The formation process begins with a random network. A subset of neurons, called training neurons, intermittently receive superthreshold external input. Gradually, a synfire chain emerges through a recruiting process, in which neurons within the network connect to the tail of the chain started by the training neurons. The model is robust to varying parameters, as well as natural events like neuronal turnover and massive lesions. Our model suggests that long synfire chain can form during the development through self-organization, and axon remodeling, ubiquitous in developing neural circuits, is essential in the process.     

Evolution with material symbol systems

Pattee's semantic closure principle is used to study the characteristics and requirements of evolving material symbols systems. By contrasting agents that reproduce via genetic variation with agents that reproduce via self-inspection, we reach the conclusion that symbols are necessary to attain open-ended evolution, but only if the phenotypes of agents are the result of a material, self-organization process. This way, a study of the inter-dependencies of symbol and matter is presented. This study is based first on a theoretical treatment of symbolic representations, and secondly on simulations of simple agents with matter-symbol inter-dependencies. The agent-based simulations use evolutionary algorithms with indirectly encoded phenotypes. The indirect encoding is based on Fuzzy Development programs, which are procedures for combining fuzzy sets in such a way as to model self-organizing development processes.     

Where’s the Noise? Key Features of Spontaneous Activity and Neural Variability Arise through Learning in a Deterministic Network

Even in the absence of sensory stimulation the brain is spontaneously active. This background “noise” seems to be the dominant cause of the notoriously high trial-to-trial variability of neural recordings. Recent experimental observations have extended our knowledge of trial-to-trial variability and spontaneous activity in several directions: 1. Trial-to-trial variability systematically decreases following the onset of a sensory stimulus or the start of a motor act. 2. Spontaneous activity states in sensory cortex outline the region of evoked sensory responses. 3. Across development, spontaneous activity aligns itself with typical evoked activity patterns. 4. The spontaneous brain activity prior to the presentation of an ambiguous stimulus predicts how the stimulus will be interpreted. At present it is unclear how these observations relate to each other and how they arise in cortical circuits. Here we demonstrate that all of these phenomena can be accounted for by a deterministic self-organizing recurrent neural network model (SORN), which learns a predictive model of its sensory environment. The SORN comprises recurrently coupled populations of excitatory and inhibitory threshold units and learns via a combination of spike-timing dependent plasticity (STDP) and homeostatic plasticity mechanisms. Similar to balanced network architectures, units in the network show irregular activity and variable responses to inputs. Additionally, however, the SORN exhibits sequence learning abilities matching recent findings from visual cortex and the network’s spontaneous activity reproduces the experimental findings mentioned above. Intriguingly, the network’s behaviour is reminiscent of sampling-based probabilistic inference, suggesting that correlates of sampling-based inference can develop from the interaction of STDP and homeostasis in deterministic networks. We conclude that key observations on spontaneous brain activity and the variability of neural responses can be accounted for by a simple deterministic recurrent neural network which learns a predictive model of its sensory environment via a combination of generic neural plasticity mechanisms.     

A self-organizing neural network with a function of associative memory: Feedback-type cognitron

Previously, one of the authors proposed a new hypothesis on the organization of synaptic connections, and constructed a model of self-organizing multi-layered neural network cognitron (Fukushima, 1975). the cognitron consists of a number of neural layers with similar structure connected in a cascade one after another. We have modified the structure of the cognitron, and have developed a new network having an ability of associative memory. The new network, named a feedback-type cognitron, has not only the feedforward connections as in the conventional cognitron, but also modifiable feedback connections from the last-layer cells to the front-layer ones. This network has been simulated on a digital computer. If several stimulus patterns are repeatedly presented to the network, the interconnections between the cells are gradually organized. The feedback connections, as well as the conventional feedforward ones, are self-organized depending on the characteristies of the externally presented stimulus patterns. After adequate number of stimulus presentations, each cell usually acquires the selective responsiveness to one of the stimulus patterns which have been frequently given. That is, every different stimulus pattern becomes to elicit an individual response to the network. After the completion of the self-organization, several stimulus patterns are presented to the network, and the responses are observed. Once a stimulus is given to the network, the signal keeps circulating in the network even after cutting off the stimulus, and the response gradually changes. Even though an imperfect or an ambiguous pattern is presented, the response usually converges to one of the patterns which have been frequently given during the process of self-organization. In some cases, however, a new pattern which has never been presented before, emerges. It is seen that this feedback-type cognitron has characteristics quite similar to some functions of the brain, such as the associative recall of memory, or the creation of a new idea by intuition.     

Randomness in the heparin polymer: computer simulations of alternative action patterns of heparin lyase

Heparin is a mixture of linear polysaccharides of undetermined sequence. Both biosynthetic data and computer simulation studies have established that each heparin polymer chain is comprised of oligosaccharides of defined sequence, representing ordered domains. One such ordered domian is a pentasaccharide corresponding to heparin's antithrombin III binding site. Previous computer simulation studies, performed under the assumption that heparin lyase (heparinase, EC 4.2.2.7), has a random endolytic action pattern, suggested that certain of these ordered oligosaccharide domains may themselves be nonrandomly arranged in the heparin polymer. The present work presents computer simulations of alternative action patterns for heparin lyase while assuming a random distribution of these oligosaccharide units within the heparin polymer. We consider action patterns that are determined solely by the primary structure of the substrate molecules. Results of the simulations are compared to (1) the experimental measurements of product chains formed throughout the reaction and (2) the change in weight average molecular weight Mw as a function of reaction completion as determined by absorbance at 232 nm. From the simulation of 60 action patterns for heparin lyase, we infer that one of the following statements concerning heparin and heparin lyase is true: (1) Heparin is a random arrangement of a small number of structurally defined oligosaccharide units. Heparin lyase changes its action pattern during the depolymerization of heparin (perhaps influenced by the secondary structure of substrate). (2) Heparin contain clusters of oligosaccharide sequences that are present in low concentrations (overall) in the polymer. Heparin lyase has a specificity for cleaving glycosidic linkages either exolytically at the nonreducing terminus of a chain or (endolytically) at the reducing side of these rare oligosaccharide sequence.     

A high-order graph generating self-organizing structure

A large class of neural network models have their units organized in a lattice with fixed topology or generate their topology during the learning process. These network models can be used as neighborhood preserving map of the input manifold, but such a structure is difficult to manage since these maps are graphs with a number of nodes that is just one or two orders of magnitude less than the number of input points (i.e., the complexity of the map is comparable with the complexity of the manifold) and some hierarchical algorithms were proposed in order to obtain a high-level abstraction of these structures. In this paper a general structure capable to extract high order information from the graph generated by a large class of self-organizing networks is presented. This algorithm will allow to build a two layers hierarchical structure starting from the results obtained by using the suitable neural network for the distribution of the input data. Moreover the proposed algorithm is also capable to build a topology preserving map if it is trained using a graph that is also a topology preserving map.