Bringing Anatomical Information into Neuronal Network Models.

For constructing neuronal network models computational neuroscientists have access to wide-ranging anatomical data that nevertheless tend to cover only a fraction of the parameters to be determined. Finding and interpreting the most relevant data, estimating missing values, and combining the data and estimates from various sources into a coherent whole is a daunting task. With this chapter we aim to provide guidance to modelers by describing the main types of anatomical data that may be useful for informing neuronal network models. We further discuss aspects of the underlying experimental techniques relevant to the interpretation of the data, list particularly comprehensive data sets, and describe methods for filling in the gaps in the experimental data. Such methods of "predictive connectomics" estimate connectivity where the data are lacking based on statistical relationships with known quantities. Exploiting organizational principles that link the plethora of data in a unifying framework can be useful for informing computational models. Besides overarching principles, we touch upon the most prominent features of brain organization that are likely to influence predicted neuronal network dynamics, with a focus on the mammalian cerebral cortex. Given the still existing need for modelers to navigate a complex data landscape full of holes and stumbling blocks, it is vital that the field of neuroanatomy is moving toward increasingly systematic data collection, representation, and publication.

Neural associative memories for the integration of language, vision and action in an autonomous agent.

Language understanding is a long-standing problem in computer science. However, the human brain is capable of processing complex languages with seemingly no difficulties. This paper shows a model for language understanding using biologically plausible neural networks composed of associative memories. The model is able to deal with ambiguities on the single word and grammatical level. The language system is embedded into a robot in order to demonstrate the correct semantical understanding of the input sentences by letting the robot perform corresponding actions. For that purpose, a simple neural action planning system has been combined with neural networks for visual object recognition and visual attention control mechanisms.

Pelvicaliceal biomodeling as an aid to achieving optimal access in percutaneous nephrolithotripsy.

PURPOSE: To investigate prospectively the benefits of three-dimensional stereolithographic biomodeling produced from CT data as an aid to achieving optimal access for percutaneous nephrolithotripsy (PCNL). PATIENTS AND METHODS: Eight patients with complex urinary calculi were selected. Multislice CT scans of the kidney in native and excretory phases were acquired with the patient in the prone position to simulate the position during surgery. Contiguous reconstructed slices were produced from the data volume. The data of interest were processed to transform them into a format acceptable for production of a biomodel. Exact plastic replicas of the pelvicaliceal system and the calculi were created and used for morphologic assessment, preoperative planning, patient education, and surgical navigation. RESULTS: The survey results were based on subjective opinions rather than objective data. The biomodels enhanced the ability to visualize a patient's unique anatomy before surgery. This aided the planning and rehearsal of endourologic procedures. CONCLUSION: Although this study is only a preliminary investigation, we postulate that biomodeling has the advantage of allowing imaging data to be displayed in a physical form. In difficult cases, this technique may improve treatment, operative planning, and communication with colleagues and patients. The limitations of the technology include the manufacturing time and cost, but more accurate puncture-site selection may reduce costs by saving operating time.

Crystal structure of RNA 3'-terminal phosphate cyclase, a ubiquitous enzyme with unusual topology.

BACKGROUND: RNA cyclases are a family of RNA-modifying enzymes that are conserved in eucarya, bacteria and archaea. They catalyze the ATP-dependent conversion of the 3'-phosphate to the 2',3'-cyclic phosphodiester at the end of RNA, in a reaction involving formation of the covalent AMP-cyclase intermediate. These enzymes might be responsible for production of the cyclic phosphate RNA ends that are known to be required by many RNA ligases in both prokaryotes and eukaryotes. RESULTS: The high-resolution structure of the Escherichia coli RNA 3'-terminal phosphate cyclase was determined using multiwavelength anomalous diffraction. Two orthorhombic crystal forms of E. coli cyclase (space group P2(1)2(1)2(1) and P2(1)2(1)2) were used to solve and refine the structure to 2.1 A resolution (R factor 20.4%; R(free) 27.6%). Each molecule of RNA cyclase consists of two domains. The larger domain contains three repeats of a folding unit comprising two parallel alpha helices and a four-stranded beta sheet; this fold was previously identified in translation initiation factor 3 (IF3). The large domain is similar to one of the two domains of 5-enolpyruvylshikimate-3-phosphate synthase and UDP-N-acetylglucosamine enolpyruvyl transferase. The smaller domain uses a similar secondary structure element with different topology, observed in many other proteins such as thioredoxin. CONCLUSIONS: The fold of RNA cyclase consists of known elements connected in a new and unique manner. Although the active site of this enzyme could not be unambiguously assigned, it can be mapped to a region surrounding His309, an adenylate acceptor, in which a number of amino acids are highly conserved in the enzyme from different sources. The structure of E. coli cyclase will be useful for interpretation of structural and mechanistic features of this and other related enzymes.

A new class of multi-stable neural networks: stability analysis and learning process.

Recently, multi-stable Neural Networks (NN) with exponential number of attractors have been presented and analyzed theoretically; however, the learning process of the parameters of these systems while considering stability conditions and specifications of real world problems has not been studied. In this paper, a new class of multi-stable NNs using sinusoidal dynamics with exponential number of attractors is introduced. The sufficient conditions for multi-stability of the proposed system are posed using Lyapunov theorem. In comparison to the other methods in this class of multi-stable NNs, the proposed method is used as a classifier by applying a learning process with respect to the topological information of data and conditions of Lyapunov multi-stability. The proposed NN is applied on both synthetic and real world datasets with an accuracy comparable to classical classifiers.

A gene fusion system for generating antibodies against short peptides.

A novel method to obtain specific antibodies against short peptides is described, involving synthesis of the corresponding oligodeoxynucleotides followed by cloning into a new set of fusion vectors, pEZZ8 and pEZZ18, based on two synthetic IgG-binding domains (ZZ) of Staphylococcus aureus protein A. The soluble gene fusion product thus obtained, can be collected from the culture medium of Escherichia coli and rapidly recovered in a one-step procedure by IgG affinity chromatography. The system was used to express a fusion protein consisting of the two Z fragments and the C-terminal part [amino acids (aa) 57-70] of human insulin-like growth factor I (IGF-I). This 16-kDa protein was purified by affinity chromatography on IgG Sepharose and antibodies were raised in rabbits. The fusion protein elicited peptide-specific antibodies, as measured by solid-phase radioimmuno assay and Western blotting, reactive with both synthetic C-terminal peptide and the native human IGF-I protein. The results suggests that the gene fusion system can be used for efficient antibody production against short peptides encoded by synthetic oligodeoxynucleotides.

A covalently bound catalytic intermediate in Escherichia coli asparaginase: crystal structure of a Thr-89-Val mutant.

Escherichia coli asparaginase II catalyzes the hydrolysis of L-asparagine to L-aspartate via a threonine-bound acyl-enzyme intermediate. A nearly inactive mutant in which one of the active site threonines, Thr-89, was replaced by valine was constructed, expressed, and crystallized. Its structure, solved at 2.2 A resolution, shows high overall similarity to the wild-type enzyme, but an aspartyl moiety is covalently bound to Thr-12, resembling a reaction intermediate. Kinetic analysis confirms the deacylation deficiency, which is also explained on a structural basis. The previously identified oxyanion hole is described in more detail.

Density of neurons and synapses in the cerebral cortex of the mouse.

Quantitative anatomical investigations provide the basis for functional models. In this study the density of neurons and synapses was measured in three different areas (8, 6, and 17) of the neocortex of the mouse. Both kinds of measurements were made on the same material, embedded in Epon/Araldit. In order to determine the synaptic density per mm3, the proportion of synaptic neuropil was also measured; it was found to be 84%. The cortical volume occupied by cell bodies of neurons and glia cells amounted to 12%, that by blood vessels to 4%. The total average was 9.2 x 10(4) neurons/mm3 and 7.2 x 10(8) synapses/mm3. About 11% of the synapses were of type II. The density of neurons increased with decreasing cortical thickness; thus the number of neurons under a given surface area was about constant. The synaptic density, on the other hand, was almost constant in the three areas, the number of synapses under a given cortical surface area tended, therefore, to increase with cortical thickness. The average number of synapses per neuron was 8,200, with a tendency to increase with increasing cortical thickness. Shrinkage of the tissue was also measured for various staining techniques. No shrinkage occurred during perfusion with 3.7% formaldehyde or with a solution of buffered paraformaldehyde and glutaraldehyde and during fixation in situ. Electron microscopical material showed almost no shrinkage, whereas Nissl-preparations on paraffin-embedded material had only 43% of their original volume. After Nissl stain on frozen sections the volume had shrunken to 68% and after Golgi impregnation and embedding in celloidin to 70%. The total volume of the neocortex was 112 mm3 (both hemispheres together). The total number of neurons was thus 1.0 x 10(7) and the total number of synapses 8.1 x 10(10).

Calculation of surface areas from serial sections.

A new method is introduced to determine the surface area of an anatomical structure from serial sections. For highly folded surfaces the new method is more adequate than previously used methods.

Disappearance of neonatal seizures and low CSF GABA levels after treatment with vitamin B6.

In an infant with neonatal seizures, CSF GABA levels were determined before and after treatment with vitamin B6. Before onset of treatment, the level of GABA in CSF was very low (13 pmol/ml). Injection of vitamin B6 blocked the seizures immediately. When GABA level in CSF was again analysed after continued treatment with vitamin B6, a value of 127 pmol/ml was determined, which is within the normal concentration range in children. The data substantiate previous findings in brain tissue from a patient with vitamin B6-dependent seizures, and strongly indicate that impairment of central GABAergic activity was the cause of the seizures.

Pattern separation and synchronization in spiking associative memories and visual areas.

Scene analysis in the mammalian visual system, conceived as a distributed and parallel process, faces the so-called binding problem. As a possible solution, the temporal correlation hypothesis has been suggested and implemented in phase-coding models. We propose an alternative model that reproduces experimental findings of synchronized and desynchronized fast oscillations more closely. This model is based on technical considerations concerning improved pattern separation in associative memories on the one hand, and on known properties of the visual cortex on the other. It consists of two reciprocally connected areas, one corresponding to a peripheral visual area (P), the other a central association area (C). P implements the orientation-selective subsystem of the primary visual cortex, while C was modeled as an associative memory with connections formed by Hebbian learning of all assemblies corresponding to stimulus objects. Spiking neurons including habituation and correlated noise were incorporated as well as realistic synaptic delays. Three learned stimuli were presented simultaneously and correlation analysis was performed on spike recordings. Generally, we found two states of activity: (i) relatively slow and unordered oscillations at about 20-25 Hz, synchronized only within small regions; and (ii) faster and more precise oscillations around 50-60 Hz, synchronized over the whole simulated area. The neuron groups representing one stimulus tended to be simultaneously in either the slow or the fast state. At each particular time, only one assembly was found to be in the fast state. Activation of the three assemblies switched on a time scale of 100 ms. This can be interpreted as self-generated attention switching. On the time scale corresponding to gamma oscillations, cross correlations between local neuron groups were either modulated or flat. Modulated correlograms resulted if the groups coded features corresponding to a common object. Otherwise, the correlograms remained flat. This behavior is in agreement with experimental results, while phase-code models would generally predict modulated correlations also in the case of different objects. Furthermore, we derive a technical version from our biological associative memory model that accomplishes fast pattern separation parallel in O(log2 n) steps for n neurons and sparse coding.

Three learning phases for radial-basis-function networks.

In this paper, learning algorithms for radial basis function (RBF) networks are discussed. Whereas multilayer perceptrons (MLP) are typically trained with backpropagation algorithms, starting the training procedure with a random initialization of the MLP's parameters, an RBF network may be trained in many different ways. We categorize these RBF training methods into one-, two-, and three-phase learning schemes. Two-phase RBF learning is a very common learning scheme. The two layers of an RBF network are learnt separately; first the RBF layer is trained, including the adaptation of centers and scaling parameters, and then the weights of the output layer are adapted. RBF centers may be trained by clustering, vector quantization and classification tree algorithms, and the output layer by supervised learning (through gradient descent or pseudo inverse solution). Results from numerical experiments of RBF classifiers trained by two-phase learning are presented in three completely different pattern recognition applications: (a) the classification of 3D visual objects; (b) the recognition hand-written digits (2D objects); and (c) the categorization of high-resolution electrocardiograms given as a time series (ID objects) and as a set of features extracted from these time series. In these applications, it can be observed that the performance of RBF classifiers trained with two-phase learning can be improved through a third backpropagation-like training phase of the RBF network, adapting the whole set of parameters (RBF centers, scaling parameters, and output layer weights) simultaneously. This, we call three-phase learning in RBF networks. A practical advantage of two- and three-phase learning in RBF networks is the possibility to use unlabeled training data for the first training phase. Support vector (SV) learning in RBF networks is a different learning approach. SV learning can be considered, in this context of learning, as a special type of one-phase learning, where only the output layer weights of the RBF network are calculated, and the RBF centers are restricted to be a subset of the training data. Numerical experiments with several classifier schemes including k-nearest-neighbor, learning vector quantization and RBF classifiers trained through two-phase, three-phase and support vector learning are given. The performance of the RBF classifiers trained through SV learning and three-phase learning are superior to the results of two-phase learning, but SV learning often leads to complex network structures, since the number of support vectors is not a small fraction of the total number of data points.

Evidence, information, and surprise.

A numerical measure for "evidence" is defined in a probabilistic framework. The established mathematical concept of information or entropy (as defined in ergodic theory) can be obtained from this definition in a special case, although in general information in a special case, although in general information is greater than evidence. In another, somewhat complementary, special case a numerical measure for "surprise" is derived from the definition of evidence. Some applications of the new concept of evidence are discussed, concerning statistics in general and the special kind of statistics performed by neurophysiologists, when they analyze the "response" of neurons, and perhaps by the neurons themselves.

Towards a theory of cell assemblies.

The term cell assembly, first introduced by D. O. Hebb, is defined in the framework of graph theory. This definition leads to some beautiful problems concerning the number and size of cell assemblies in large graphs. Some approaches to solve these problems are presented. In particular, the graphs Kn X Km are constructed that have n . m points, n + m - 2 connections per point, and at least 2n + 2m - 4 assemblies. Several new notions of connectivity in directed graphs are introduced and their relationships are investigated. The insight into these notions and their relationships will be helpful for further construction of graphs with many assemblies and/or high connectivity. The resulting graphs are not only important for the idea of cell assemblies in the content of neurodynamics, they may also find applications in the construction of communication networks and associative memories.

On associative memory.

The information storing capacity of certain associative and auto-associative memories is calculated. For example, in a 100 X 100 matrix of 1 bit storage elements more than 6,500 bits can be stored associatively, and more than 688,000 bits in a 1,000 X 1,000 matrix. Asymptotically, the storage capacity of an associative memory increases proportionally to the number of storage elements. The usefulness of associative memories, as opposed to conventional listing memories, is duscussed -- especially in connection with brain modelling.

Parallel processing for associative and neuronal networks.

We represent a parallel processing network, consisting of nine microcomputers, for neuron-network simulations and for the realization of an associative computer memory. We add some remarks on the present possibilities to implement larger associative networks and on parallel processing strategies in general.

Perceptual rivalry between illusory and real contours.

The interactions between illusory and real contours have been investigated under monocular, binocular and dichoptic conditions. Results show that under all three presentation conditions, periodic alterations, generally called rivalry, occur during the perception of cognitive (or illusory) triangles, while earlier research had failed to find such rivalry (Bradley and Dumais 1975). With line triangles, rivalry is experienced only under dichoptic conditions. A model is proposed to account for the observed phenomena.