Summary Computational Neuroscience
Lecture 1, Introduction to Computational Neuroscience, Neuron
Models
What is computational neuroscience?
Computational neuroscience is a branch of neuroscience which employs mathematical
models, computer simulations, theoretical analysis and abstractions of the brain to
understand the principles that govern the development, structure, physiology and cognitive
abilities of the nervous system
Computational neuroscience is also known as theoretical neuroscience or mathematical
neuroscience
Making simulations to see if it will have the same behaviour
Modelling the Brain: Neuron models
• Why do we need mathematical models?
What I cannot create I do not understand
• How do we model the brain?
• Neuron models
• Leaky Integrate and Fire model
• Hodgkin-Huxley (HH) model
• Multi-compartment HH models
• Human Brain project
• Building detailed network models
Why do we need mathematical models in neuroscience?
-” Word models” (word models are not quantitative and if it is not quantitative you can make
all kind of scenarios, the interactions are not confined) sound reasonable but equations force
a model to be precise, complete, and self-consistent (the most difficult part to make it
reproducible to reality)
-Fast way to generate and vet ideas prior to full experimental testing. (no unnecessary
experiments (animals); you want the optimal sensitivity; saves time too)
-The key test of the value of a theory is not necessarily whether it predicts something new,
but whether it produces concepts that generalize to other systems and provide valuable new
ways of thinking. (we want to understand concepts because those things lead to
understanding questions of life)
,Brain-inspired computing: low energy computing
???
You have to show a computer a bicycle or scooter more times than a human who only has to
see this one or twice to learn this.
- You want very fast but low-energy
Brain-inspired computing: Artificial intelligence
“deep learning” inspired AI
- Very big network with high-speed
- Google and Siri etc. all is trained by a network with input and output
- Is there a difference between AI and human brains?
For a lot of proteins, the 3D structure is not known yet, deep learning with primary proteins
3D structures that are known to see if they can predict 3D structures of other proteins.
,What is a good model? → accurate predictions that can be tested experimentally> If they are
true it is a good model
-Just enough complexity to understand the phenomena you want to describe (not too many
(then you may not understand the system) and not too few variables depending on the
question you want to answer→ makes it realistic; but what is the minimal amount of
variables)
-if it produces emerging properties, properties of the system that are not explicitly modelled
in the individual components but arise from interaction of the different elements
In 1 individual water molecule, no property will make a snowflake, but you can still explain it
due to the interaction of multiple water molecules. Snowflake crystals are an emerging
property of water molecules.
Another example: potassium, sodium and other properties of an action potential make
together the emerging property of an action potential.
What biological aspects do we need to model if we want to simulate the brain?
- temporal and spatial scale
Depends on what aspect you want to model, choose the right level of complexity.
-molecular level
-cellular level
-network level
-systems level
The challenge is to connect these levels
Brain models:
-Excitability (AP generation)
-Synaptic transmission and plasticity
-AP propagation
-Connectivity
-Morphology
-memory
-information processing
-brain disease
, Linking different scales in the brain through computational modelling
• Properties and interactions of components at a lower level determine the behavior
(emergent properties) of the system one level up.
• For example: the electrical properties of individual neurons and their synaptic connections
determine the frequency of oscillations (emergent property) observed in neuronal networks.
• With this approach we can for instance try to understand how patient mutations lead to
brain disease. E.g. mutations in synaptic genes will affect protein function, altered protein
function will affect synapse function, altered synapse function will affect network behavior,
etc.
Coarse-grained molecular simulations
3-microsecond simulation of SNARE complex-driven fusion pore formation
- They try to simulate how an NT can be released through a pore when a AP arrives at a
synapse
- It is not feasible to simulate the whole brain at this level, but it gives an emerging
property, if you have the molecules, you can do this fusion. A mutation in one of the
properties has big impact and might affect signalling in a network
- The emerging property is that you can see NT release due to SNARE proteins
Lecture 1, Introduction to Computational Neuroscience, Neuron
Models
What is computational neuroscience?
Computational neuroscience is a branch of neuroscience which employs mathematical
models, computer simulations, theoretical analysis and abstractions of the brain to
understand the principles that govern the development, structure, physiology and cognitive
abilities of the nervous system
Computational neuroscience is also known as theoretical neuroscience or mathematical
neuroscience
Making simulations to see if it will have the same behaviour
Modelling the Brain: Neuron models
• Why do we need mathematical models?
What I cannot create I do not understand
• How do we model the brain?
• Neuron models
• Leaky Integrate and Fire model
• Hodgkin-Huxley (HH) model
• Multi-compartment HH models
• Human Brain project
• Building detailed network models
Why do we need mathematical models in neuroscience?
-” Word models” (word models are not quantitative and if it is not quantitative you can make
all kind of scenarios, the interactions are not confined) sound reasonable but equations force
a model to be precise, complete, and self-consistent (the most difficult part to make it
reproducible to reality)
-Fast way to generate and vet ideas prior to full experimental testing. (no unnecessary
experiments (animals); you want the optimal sensitivity; saves time too)
-The key test of the value of a theory is not necessarily whether it predicts something new,
but whether it produces concepts that generalize to other systems and provide valuable new
ways of thinking. (we want to understand concepts because those things lead to
understanding questions of life)
,Brain-inspired computing: low energy computing
???
You have to show a computer a bicycle or scooter more times than a human who only has to
see this one or twice to learn this.
- You want very fast but low-energy
Brain-inspired computing: Artificial intelligence
“deep learning” inspired AI
- Very big network with high-speed
- Google and Siri etc. all is trained by a network with input and output
- Is there a difference between AI and human brains?
For a lot of proteins, the 3D structure is not known yet, deep learning with primary proteins
3D structures that are known to see if they can predict 3D structures of other proteins.
,What is a good model? → accurate predictions that can be tested experimentally> If they are
true it is a good model
-Just enough complexity to understand the phenomena you want to describe (not too many
(then you may not understand the system) and not too few variables depending on the
question you want to answer→ makes it realistic; but what is the minimal amount of
variables)
-if it produces emerging properties, properties of the system that are not explicitly modelled
in the individual components but arise from interaction of the different elements
In 1 individual water molecule, no property will make a snowflake, but you can still explain it
due to the interaction of multiple water molecules. Snowflake crystals are an emerging
property of water molecules.
Another example: potassium, sodium and other properties of an action potential make
together the emerging property of an action potential.
What biological aspects do we need to model if we want to simulate the brain?
- temporal and spatial scale
Depends on what aspect you want to model, choose the right level of complexity.
-molecular level
-cellular level
-network level
-systems level
The challenge is to connect these levels
Brain models:
-Excitability (AP generation)
-Synaptic transmission and plasticity
-AP propagation
-Connectivity
-Morphology
-memory
-information processing
-brain disease
, Linking different scales in the brain through computational modelling
• Properties and interactions of components at a lower level determine the behavior
(emergent properties) of the system one level up.
• For example: the electrical properties of individual neurons and their synaptic connections
determine the frequency of oscillations (emergent property) observed in neuronal networks.
• With this approach we can for instance try to understand how patient mutations lead to
brain disease. E.g. mutations in synaptic genes will affect protein function, altered protein
function will affect synapse function, altered synapse function will affect network behavior,
etc.
Coarse-grained molecular simulations
3-microsecond simulation of SNARE complex-driven fusion pore formation
- They try to simulate how an NT can be released through a pore when a AP arrives at a
synapse
- It is not feasible to simulate the whole brain at this level, but it gives an emerging
property, if you have the molecules, you can do this fusion. A mutation in one of the
properties has big impact and might affect signalling in a network
- The emerging property is that you can see NT release due to SNARE proteins
