Tutorial:Single Channel Kinetics

From QuB

Jump to: navigation, search
Prev: Single Channel Kinetics Outline Next: Single Channel Visualizations


Contents

Basic MIL

This tutorial covers the basic use of MIL to build a model and solve for rate constants, and the effects of missed events correction.

c-o
c-o
c-o-c
c-o-c
c-o-c -c
c-o-c -c
  1. If you haven't already, simulate basic model sim.qdf
  2. Idealize it (steps 1-7)
  3. Build the model "c-o" at right
  4. Set the dead time: click "td" at the upper right until it says "td [samples]". Enter a value of 1.0, meaning that events of 1 sample duration or shorter may have been missed by the idealization.
  5. Right-click "MIL" on the right under "Modeling"
  6. Choose the following options:
    Data source Whole file
    Run mode Optimize ("check model" would calculate LL and histograms without optimizing rates)
    Use segments Together ("separately" would find different optimal rates for each data segment)
  7. Click "Run"
  8. In the Report window (View -> Report), look at the error limits on the rate constants (k0.std)
  9. In the Results window (View -> Results), observe how poorly the PDF fits the first (closed-time) histogram. There are two peaks so we will need two closed states.
  10. In the Results window, Summary tab, observe that "Gradient" is far from 0. Take note of "LL per event" for comparison with the next model. Make sure "ErrorCode" is 0, or the results could be invalid.
  11. Add a state, to produce the model "c-o-c" at right. Don't worry if your rate constants are a little different.
  12. Left-click MIL to run it with the same settings.
  13. Look at the histogram fit. Is it better? Good enough?
  14. Look at the summary. Are the gradients smaller or zero? Is the LL per event larger?
  15. Are the rate constants close to the ones we used to simulate the data?
  16. To see the effect of missed-event correction, change the dead time to 2.0 samples and run MIL. Look at the histograms, summary, and rate constants.
  17. Change the dead time to 0.5 and run MIL. This one should have the most accurate rates, best histogram fit, and smallest gradients. 0.5 samples is an appropriate dead time because the simulator samples perfectly and we idealized without a filter -- events one sample long are detected.
  18. Make a report for your records (MS Office 2000 or newer required): In the Results window, click Reports and choose "MIL Word Report"
  19. The histograms and gradients should convince you that three states are enough, but as an exercise add another closed state (c-o-c -c at right) and run MIL. Observe that LL per event hasn't improved and remove the state. Also try adding another open state.


MIL Variations

Try these variations on the tutorial above:

Reduce the number of events
Simulate the same data with fewer events and run MIL. The accuracy of the rate constants is decreased.
Reduced signal to noise ratio
Decrease the amplitude of class 2 (red), then re-simulate and run MIL
Noisy data
Increase the std. deviation of both classes (black and red), simulate, idealize, MIL. Try idealizing with the filter enabled. (check "Fc" in the top toolbar, idealize with "process data: as displayed")
Multiple conductance states
Simulate with a three-state model with each state having a different color and current amplitude. Idealize and MIL with the same model.
Multiple channels
Simulate n identical channels by setting "Channel count" to n. In principle you can analyze data from patches with several channels, but SKM and MIL will run much more slowly as n increases, so try n=2 or 3. Likewise, the number of conductance classes does not matter, but for simplicity you may wish to use a model with a single open conductance class. Idealize and MIL with the same model. The MIL outputs will consist of a histogram and time constants for each conductance level in the record, but the rate constants should pertain to the simulated rate constants for one channel.
Adding unneeded connections to a model
Note that the LL may increase, because you have increased the number of free parameters, but the rates will indicate that the transition almost never occurs.

Auto-MIL

The procedure in the Basic MIL tutorial has been automated in the script named Star. There is also a variant named Chain which builds linear models.

  1. If you haven't already, simulate basic model sim.qdf
  2. Idealize it (steps 1-7)
  3. Build the model "c-o" (above)
  4. Set the dead time to 0.5 samples
  5. Click Star under Scripts on the right; or from the menu bar, Actions -> Scripts -> MakeStar
  6. Enter "Delta LL:" 10.0, "Starting state:" 2 (the red one)
  7. Click "Run"

Be aware that Star will work as well with any starting state:

  1. In the Results window, Summary tab, note the LL from Star
  2. Make a new "c-o" model
  3. Run Star with "Starting state:" 1 (the black one)
  4. Note the new LL

Detailed Balance

This tutorial shows how to impose detailed balance on a model. We will simulate data from a model that is out of balance, then compare constrained and unconstrained models using MIL's LL.

  1. Build the model at right
  2. Simulate
  3. Idealize
  4. Add a loop-balance constraint:
    • Right-click the background of the model and choose Properties
    • Pick the tab "Kinetic constraints"
    • Click "Balance Loops"
    • Click "OK"
  5. Run MIL
  6. Look at the LL (Results window, Summary tab)
  7. Remove the constraint:
    • Model Properties, Kinetic constraints
    • Click "Delete", "OK"
  8. Run MIL
  9. Look at the LL. It should be bigger, indicating that this data is not in detailed balance.

Global Fitting

This tutorial shows how to find rate constants for a ligand-dependent model, using data simulated at different concentrations. The same procedure applies for voltage-dependent models.

  1. If you haven't already, simulate the files in the Ligand-dependent simulation tutorial
  2. Make sure all three files are open and idealized
  3. Build the model at right. To add ligand-dependence:
    • Double-click one of the dependent rates (labeled with "p" at right)
    • Check "P" and make sure the label is "Ligand"
    • Click the other dependent rate and do the same
    • Click "OK"
  4. Run MIL:
    • Right-click MIL
    • Choose Data source: File list
    • Put check-marks in the "Use" column next to the three files
    • Make sure the values in the "Ligand" column are correct (0.1, 1.0, 3.0)
    • Click "Run"
  5. In the Results window, Histograms tab, scroll through each file's histograms
  6. The rate constants in the model may be displayed as k = P * k0 * eQ * k1, with P and Q values from the current data file. To read k0, double-click the rate.

Non-stationary Data

Some models are not distinguishable in equilibrium conditions (e.g., the starting probability of every state is 0), for example C-O-C and C-C-O (try this). To distinguish between two similar models, simulate a non-stationary experiment (i.e., a pulse of agonist onto a single channel, which begins in the closed state) using both models (C-O-C and C-C-O):

  1. Create 2 models: C-O-C and C-C-O.
  2. For each model, go to the ‘State and class properties’ window for the first class 0 state and set the starting probability (Starting prob) to 1. This insures that each segment begins in class one, as if an agonist pulse is being applied a resting ligand-gated channel.
  3. Simulate many short segments (so that there are only one or two openings per segment) for each model.
  4. Analyze each data set separately and notice that you will be able to distinguish the models using MIL.

Note: The dwell time histograms generated by MIL are not relevant when the currents are not in equilibrium, as in the above simulations. That is, the theoretical curves drawn for the rate constants will not fit the binned interval durations. Histograms should be ignored when the data do not represent equilibrium conditions.

Note: At the present, MIL has a limited ability to analyze non-stationary data, but full analysis is possible with MIP.

Heterogeneous Bursts

This tutorial shows how to handle bursts (clusters) of data when there is more than one kind of channel in a patch. We will

  1. Determine Tcrit, the minimum inter-burst closed duration
  2. Chop idealized data into bursts
  3. Divide the bursts into high-PO and low-PO lists.
  4. Solve the rate constants of each list.

If you haven't already, simulate heterogeneous.qdf.

  1. We define a burst as activity with closed events shorter than Tcrit (ms). Choosing Tcrit is somewhat arbitrary. The duration distributions for a short-lived closed state and a long-lived closed state overlap, so some events from each will be misclassified. We will use the Tcrit computed by MIL, which equalizes the area under the overlapping tails of the distributions.
    • Build a C-O-C model
    • "Grab" amplitude levels:
      • select a stretch of single-channel activity
      • right-click the model background, "Grab all amps increasing"
    • Set Channel count to 1
    • Set Data source: File
    • Idealize with SKM
    • Run MIL
    • The red line on the histogram is Tcrit; its value is in the Summary tab
    • Now idealize for real:
      • Set Channel count to 2
      • Idealize with SKM
  2. Click the ChopIdl button (under Preprocessing on the right)
    • For Burst terminator, choose "Dwell in class 1 longer than" Tcrit
    • Check "Dwell(s) in class" 3 or greater to exclude overlapping bursts
    • Check "Fewer than" 10 events to exclude bursts that are too short
    • Check "Compute stats" to create a Result with Popen etc.
    • Click OK
    • There should be a list of "Chopped Events" in the data window.
    • In the Data window, under Results, right-click "ChopIdl" and choose "Hold Results" to preserve the stats for the life of the data file.
  3. Use the Results window to explore and separate the two kinds of activity
    • Choose the Select tab
    • Click "Show Var..." and choose "occupancy 2". This is the occupancy probability of the open class, a.k.a Popen. A chart should appear showing selection number v. occupancy 2
    • Right-click one of the highest datapoints and "Show segment". A burst with high Popen will be highlighted in the data window. Try showing other segments. Segments with intermediate Popen are a mix of both kinds of activity.
    • Classify the selections statistically with the K-means algorithm:
      • Click Select and choose "None"
      • Click the color red, then click a representative high-Popen point to color it red.
      • Click the color blue, then click a representative low-Popen point
      • Click Select and choose "K-means x5"
      • Click Select and choose "Trim" to unselect points outside the standard-deviation lines.
      • Right-click the color red and choose "Make selection list...". Name it "High Popen"
      • Right-click the color blue and choose "Make selection list...". Name it "Low Popen"
  4. Now that we have the two kinds of bursts in separate lists, we can solve the rate constants for each list. The data was simulated with a three-state model but we've excluded events from the long closed state, leaving bursts that can be modeled with a two-state model.
    • Build a C-O model.
    • Set Channel count: 1
    • Set Data Source: List
    • Select the "High Popen" list in the Data window
    • Run MIL
    • Select the "Low Popen" list
    • Run MIL

Model Search

This tutorial shows how to use Model search, which optimizes the MIL LL for all models with a certain number of closed and open states.

  1. If you haven't already, simulate basic model sim.qdf.
  2. Idealize it if necessary
  3. Create a model with two closed states and one open state. It doesn't matter how you connect them; model search will rewire the connections.
  4. Right-click "Mdl Srch" or "msrch", under Modeling on the right
  5. Set Data Source: whole file
  6. Set Trials per model: 3 (each trial starts with random rates; repeated trials guard against local optima)
  7. Run
  8. In the Results window, choose the Models tab
    • To sort, click a column header.
    • To see a model, click its name
    • Disregard models with nonzero ErrorCode.
    • Be aware that some models will report a huge LL but have impossibly large rate constants.
    • Notice that two models (C-O-C and C-C-O) are equally likely.


Prev: Single Channel Kinetics Outline Next: Single Channel Visualizations
Retrieved from "http://www.qub.buffalo.edu/wiki/index.php/Tutorial:Single_Channel_Kinetics"
Personal tools