Transcript Neural Network Toolbox - University of Sunderland
Cell Modeling in Genesis
COMM2M Harry R. Erwin, PhD University of Sunderland
Resources
• Bower and Beeman, 1998,
The Book of Genesis,
Second edition, Springer-Verlag.
Structurally Realistic Modeling
• Why?
– Biological realism allows known neural data to be used as constraints, limiting the dimensionality of the resulting search space.
– Biological relevance makes predictions testable.
– The model-building process identifies the data researchers should be collecting.
Pedagogical Reasons
• Increased chance of generating unanticipated functional insights based on emergent properties of neurons. • The more realistic the model, the more likely it will discover something new. • It is easier to make a realistic model more abstract than the reverse.
The Questions You Must Be Ready to Answer
• “What do you know now that you did not know before?” • “How can you determine its likelihood?”
Model-Building is a Process, not an End in Itself
• A good model makes it very clear what you do not know and what you need to find out.
• It is quite possible that evolution has produced a very tight relationship between structure and function.
• AI has shown that the nervous system has solved some very hard problems. This suggests the CNS may be the
only
feasible solution to producing intelligence. It is interesting that neuropil has evolved at least three times.
The Modeling Process
• Build the model first, to aid in experimental design, the choice of data to collect, and the relevance of experiments.
• Neurobiology has a massive amount of data on its hands. Yet we don’t understand the answers to simple questions like how the cortex functions.
• Modeling is a tool for dealing with the data and for storing it efficiently.
How and When to Model
• A stepwise process: – Single neurons or networks?
– Individual modeling steps
Single Neurons or Networks?
• The best place to start is with an existing model. Single cell or network?
• It depends: – Detailed single cell models show the dynamics of responses, but do not tell how the cell is activated.
– Networks use more simplified neural models • Look at your data.
• You will do both kinds of modeling in the end.
Modeling Steps
1. Establish sufficient structural detail to replicate physiological (or non-physiological) responses. (Non-physiological responses are unnatural.) 2. Simulate results from as many different recording types as possible.
3. Then begin exploring functional properties of the model. In single-cell models, this involves adding synaptic conductances. In networks, this involves synaptic activation or plasticity.
4. Interesting results appear early, during tuning.
To Simulate a Compartment
• The compartment object is used to construct neurons. – ‘create object name [options]’ • Elements form a hierarchy like directories.
• Use ‘create neutral /cell’ to create a parent element for the hierarchy.
• ‘create compartment /cell/soma’ then creates a somatic compartment for the cell.
• To delete an object: ‘delete name’
Examining and Modifying Elements
• ‘le’ lists the elements in the current level of the hierarchy.
• Elements contain fields. To see some: ‘showfield /cell/soma -all’ • To see the membrane resistance field: ‘showfield /cell/soma Rm’ • To change the current working element to location: ‘ce location’ • This location will be the default for all commands.
Moving Around
• ‘pwe’ prints the current working element • ‘pushe location’ and ‘pope’ allows to you set up a stack of elements.
• To set field values use the ‘setfield location field value’ • Multiple fields in a location can be set by repeating the ‘field value’ arguments.
Running Genesis
• Use ‘check’ to validate the simulation.
• Use ‘reset’ to go to a known initial state.
• To step the simulation, use ‘step count’ • Graphics can be added by using the XODUS library. ‘create xform location’ • xform is a default form.
• To show a form, ‘xshow location’
Graphs
• To create a graph named
voltage
‘create xgraph /data/voltage’ • This means a given form can contain multiple graphical elements.
• To hide a form: ‘xhide location’
To Plot Data
• Interelement communication is done using messages. This can be between model elements or between model and graphical elements.
• ‘addmsg /cell/soma /data/voltage PLOT Vm *volts *red’ • Source-destination-type-datafield name label-color
Running the Simulation
• ‘reset’ • ‘step 100’
To Add Another Field to the Same Graph
• ‘addmsg /cell/soma /data/voltage PLOT inject *current *blue’ • ‘showmsg’ will list the messages entering and leaving a given element.
• ‘deletemsg’ can be used to remove messages.
Adding Buttons to a Form
• ‘create xbutton /data/RESET -script reset’ • This creates a reset button in the form and attaches the ‘reset’ script to it.
• ‘create xbutton /data/RUN -script “step 100”’ • This creates a run button that commands ‘step 100’ when pressed.
Scripting Genesis
• Create a text file, ‘name.g’ • Begin with ‘//genesis’ • Cut and paste your code into it.
• ‘//’ or ‘/*…*/’ are used for comments • ‘\’ continues a command on the next line • To run it, type ‘name’ • To run genesis with the script, type ‘genesis name’ at the unix prompt.
Simulating a Soma
• Conventions: – Put your functions at the start of the script ‘function
Example Function
function print_area(length, diameter) float length, diameter float area float PI = 3.14.15926
area = PI*diameter*length echo The area is {area} end
To Call
• ‘print_area 5 5’ • Results: ‘The area is 78.5397’ • Note how variables are initialized and how {…} constructs are used to refer to variables in print statements.
• ‘echo’ prints out its arguments • “…” are optional for strings
Units
• Genesis has no implicit units. Stick to SI (MKS) units to avoid confusion.
– Ohms, farads, volts, amperes, seconds, siemens, and meters • Often, however, physiological units are used: – Kilohm, microfarad, millivolt, microampere, millisecond, millisiemen, centimeter
Cell Parameters
• • • • •
R m
—membrane resistance
C m
—membrane capacitance •
R a
—axial resistance • Specific units, independent of cell dimensions, are usually used.
R M —
units of ohm-meter 2
C M
—units of farads/meter 2
R A
—units of ohm-meter
A Squid Soma
‘float RM = 0.33333’ ‘float CM = 0.01’ ‘float RA = 0.3’ ‘float soma_1 = 30e-6’ ‘float soma_d = 30e-6’ ‘float EREST_ACT = -0.070’ ‘float Eleak = EREST_ACT + 0.0106’ ‘float ENA = 0.045’ ‘float EK = -0.082’
makecompartment
• ‘create neutral /cell’ • ‘makecompartment /cell/soma {soma_l}\ {soma_d} {Eleak}’ • ‘setfield /cell/soma inject 0.3e-9’