Computational Neurogenetic Modeling (eBook)

Dedication 6
Preface 7
Table of Contents 9
1 Computational Neurogenetic Modeling (CNGM): A Brief Introduction 13
1.1 Motivation - The Evolving Brain 13
1.2 Computational Models of the Brain 16
1.3 Brain-Gene Data, Information and Knowledge 18
1.4 CNGM: How to Integrate Neuronal and Gene Dynamics? 24
1.5 What Computational Methods to Use for CNGM? 26
1.6 About the Book 27
1.7 Summary 28
2 Organization and Functions of the Brain 29
2.1 Methods of Brain Study 30
2.2 Overall Organization of the Brain and Motor Control 35
2.3 Learning and Memory 37
2.4 Language and Other Cognitive Functions 41
2.4.1 Innate or Learned? 41
2.4.2 Neural Basis of Language 42
2.4.3 Evolution of Language, Thinking and the Language Gene 45
2.5 Neural Representation of Information 48
2.6 Perception 49
2.7 Consciousness 53
2.7.1 Neural Correlates of Sensory Awareness 53
2.7.2 Neural Correlates of Reflective Consciousness 56
2.8 Summary and Discussion 61
3 Neuro-Information Processing in the Brain 65
3.1 Generation and Transmission of Signals by Neurons 65
3.2 Learning Takes Place in Synapses: Toward the Smartness Gene 68
3.3 The Role of Spines in Learning 70
3.4 Neocortical Plasticity 73
3.4.1 Developmental Cortical Plasticity 73
3.4.2 Adult Cortical Plasticity 76
3.4.3 Insights into Cortical Plasticity via a Computational Model 78
3.5 Neural Coding: the Brain is Fast, Neurons are Slow 86
3.5.1 Ultra-Fast Visual Classification 86
3.5.2 Hypotheses About a Neural Code 89
Coding Based on Spike Timing 89
The Rate Code 89
3.6 Summary 90
4 Artificial Neural Networks (ANN) 93
4.1 General Principles 93
4.2 Models of Learning in Connectionist Systems 96
4.3 Unsupervised Learning (Self Organizing Maps - SOM) 105
4.3.1 The SOM Algorithm 105
4.3.2 SOM Output 107
Sample Distribution 107
Clustering Information 107
Visualization of Input Variables 108
Relationship Between Multiple Descriptors 108
The Connection Weights 108
Interpretation by the Fuzzy Set Theory 109
4.3.3 SOM for Brain and Gene Data Clustering 109
4.4 Supervised Learning 110
4.4.1 Multilayer Perceptron (MLP) 110
4.4.2 MLP for Brain and Gene Data Classification 111
Example 111
4.5 Spiking Neural Networks (SNN) 114
4.6 Summary 117
5 Evolving Connectionist Systems (ECOS) 119
5.1 Local Learning in ECOS 119
5.2 Evolving Fuzzy Neural Networks EFuNN 120
5.3 The Basic EFuNN Algorithm 124
5.4 DENFIS 128
5.4.1 Dynamic Takagi-Sugeno Fuzzy Inference Engine 140
5.4.2 Fuzzy Rule Set, Rule Insertion and Rule Extraction 141
5.5 Transductive Reasoning for Personalized Modeling 142
5.5.1 Weighted Data Normalization 144
5.6 ECOS for Brain and Gene Data Modeling 144
5.6.1 ECOS for EEG Data Modeling, Classification and Signal Transition Rule Extraction 144
5.6.2 ECOS for Gene Expression Profiling 146
5.7 Summary 148
6 Evolutionary Computation for Model and Feature Optimization 149
6.1 Lifelong Learning and Evolution in Biological Species: Nurture vs. Nature 149
6.2 Principles of Evolutionary Computation 150
6.3 Genetic Algorithms 150
6.4 EC for Model and Parameter Optimization 155
6.4.1 Example 155
6.5 Summary 158
7 Gene/Protein Interactions - Modeling Gene Regulatory Networks (GRN) 159
7.1 The Central Dogma of Molecular Biology 159
7.2 Gene and Protein Expression Data Analysis and Modeling 163
7.2.1 Example 165
7.3 Modeling Gene/Protein Regulatory Networks (GPRN) 167
7.4 Evolving Connectionist Systems (ECOS) for GRN Modeling 172
7.4.1 General Principles 172
7.4.2 A Case Study on a Small GRN Modeling with the Use of ECOS 173
7.5 Summary 175
8 CNGM as Integration of GPRN, ANN and Evolving Processes 177
8.1 Modeling Genetic Control of Neural Development 178
8.2 Abstract Computational Neurogenetic Model 183
8.3 Continuous Model of Gene-Protein Dynamics 187
8.4 Towards the Integration of CNGM and Bioinformatics 193
8.5 Summary 197
9 Application of CNGM to Learning and Memory 199
9.1 Rules of Synaptic Plasticity and Metaplasticity 199
9.2 Toward a GPRN of Synaptic Plasticity 207
9.3 Putative Molecular Mechanisms of Metaplasticity 215
9.4 A Simple One Protein-One Neuronal Function CNGM 218
9.5 Application to Modeling of L-LTP 220
9.6 Summary and Discussion 224
10 Applications of CNGM and Future Development 226
10.1 CNGM of Epilepsy 227
10.1.1 Genetically Caused Epilepsies 227
10.1.2 Discussion and Future Developments 230
10.2 CNGM of Schizophrenia 231
10.2.1 Neurotransmitter Systems Affected in Schizophrenia 233
10.2.2 Gene Mutations in Schizophrenia 235
10.2.3 Discussion and Future Developments 238
10.3 CNGM of Mental Retardation 239
10.3.1 Genetic Causes of Mental Retardation 240
10.3.2 Discussion and Future Developments 244
10.4 CNGM of Brain Aging and Alzheimer Disease 245
10.5 CNGM of Parkinson Disease 250
10.6 Brain-Gene Ontology 253
10.7 Summary 256
Appendix 1 258
A.1 Table of Genes and Related Brain Functions and Diseases 258
Appendix 2 268
A.2 A Brief Overview of Computational Intelligence Methods 268
A.2.1 Probabilistic and Statistical Methods 268
A.2.2 Boolean and Fuzzy Logic Models 271
A.2.3 Artificial Neural Networks 274
A.2.4 Methods of Evolutionary Computation (EC) 277
Appendix 3 278
A.3 Some Sources of Brain-Gene Data, Information, Knowledge and Computational Models 278
References 280
Index 308

"3.2 Learning Takes Place in Synapses: Toward the Smartness Gene (p. 56-57)

For major discoveries in the field of synaptic mechanisms of learning, the 2000 Nobel Prize for medicine went to the neuroscientists Eric R. Kandel and Paul Greengard. The 3rd laureate, Arvid Carlsson, got his share of the prize for discoveries of actions of neurotransmitter dopamine. At present, it is widely accepted that learning is accompanied by changes of synaptic weights in cortical neural networks (Kandel et al. 2000). Changes of synaptic weights are also called synaptic plasticity. In 1949, the Canadian psychologist Donald Hebb formulated a universal rule for these changes: "When an axon of cell A excites cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells so that As efficiency as one of the cells firing B is increased", which has been verified in many experiments and its mechanisms elucidated (Hebb 1949).

In cerebral cortex and in hippocampus of humans and animals, learning takes place in excitatory synapses formed upon dendritic spines that use glutamate as their neurotransmitter. In the regime of learning, glutamate acts on specific postsynaptic receptors, the so-called NMDA receptors (Nmethyl- D-aspartate). NMDA receptors are associated with ion channels for sodium and calcium (see Fig. 3.3). The influx of these ions into spines is proportional to the frequency of incoming presynaptic spikes. Calcium acts as a second messenger thus triggering a cascade of biochemical reactions which lead either to the long-term potentiation of synaptic weights (LTP) or to the long-term depression (weakening) of synaptic weights (LTD).

In experimental animals, it has been recorded that these changes in synaptic weights can last for hours, days, even weeks and months, up to a year. Induction of such long-term synaptic changes involves transient changes in gene expression (Mayford and Kandel 1999, Abraham et al. 2002). A subcellular switch between LTD and LTP is the concentration of calcium within spines (Shouval, Bear et al. 2002). We speak about an LTD/LTP threshold. In tum, the intra-spine calcium concentration depends upon the intensity of synaptic stimulation that is upon the frequency of presynaptic spikes.

That is, more presynaptic spikes mean more glutamate within synaptic cleft. Release of glutamate must coincide with a sufficient depolarization of the postsynaptic membrane to remove the magnesium block ofthe NMDA receptor. The greater the depolarization, the more ions of calcium enters the spine. Postsynaptic depolarization is primarily achieved via AMPA (amino-methylisoxasole-propionic acid) receptors, however, recently a significant role ofbackpropagating postsynaptic spikes has been pointed out (Markram et al. 1997). Calcium concentrations below or above the LTD/LTP threshold, switch on different enzymatic pathways that lead either to LTD or LTP, respectively. However, the current value of the LTD/LTP threshold (i.e. the properties of these two enzymatic pathways) can be influenced by levels of other neurotransmitters, an average previous activity of a neuron, and possibly other biochemical factors as well.

This phenomenon is called metaplasticity, a plasticity of synaptic plasticity (Abraham and Bear 1996). Dependence of the LTD/LTP threshold upon different postsynaptic factors is the subject of the Bienenstock, Cooper and Munro (BCM) theory of synaptic plasticity (Bienenstock et al. 1982) (for a nice overview see for instance (Jedlicka 2002)). The BCM theory of synaptic plasticity has been successfully applied in computer simulations to explain experience-dependent changes in the normal and ultrastructurally altered brain cortex of experimental animals (Benuskova et al. 1994)."