Answers

From QuB

Jump to: navigation, search
Prev: Multi-channel Popen Outline Next: Reasons to use QuB


We encourage you to ask questions on our forum, and to join the discussion. If you prefer, you can email questions to "qubhelp @ www.qub.buffalo.edu". Below are some questions we've answered by email. There are a lot more in the forum.

Contents

How to export data for Clampfit

1. Extract the recording ('Ext' button) into ASC format without checking any of the options of the 'output format' box and checking the box 'Join Segments' ('Only active channel' also checked) - I select in the window that appears next '4 bytes' and check 'Float' (although this is only to show the ASC data into QUB again). This generates a TXT file.

2. Open the TXT file using the pClamp software - the program I use is called Clampfit and the version is 9.2.0.09 (I hope all this doesn't change with the version). In the window that appears next I check 'Gap-free' and 'Add a time column', assigning the correct sampling interval value. This imports the recording into Clampfit so that we can see the same recording we had in QUB now in Clampfit's window.

3. Last, save the recording as (File/Save as... option) an ABF file

thanks to Jose Angel Fernandez

How to set Q (Voltage) 

    - how can I set the Q value when modeling voltage-dependency of rate
    constants? When I open the Rate properties window, I can check the box
    for Q, but there is no box for giving it a value.

Each data file can have its own Q value. In the rate properties, note the name next to Q (eg Voltage). Then go to the data properties, Info tab, and set the Value of Voltage. For simulations, set the Voltage in the simulation properties, model tab.

Multichannel Patches 

how can i tackle multichannel patch in QuB. i.e idealization.
lets say that i have a two channel trace. how can i calculate
the Po, dwell time and construct the amplitude histogram?

last but not the least - to be sure that my patch containing
multiple channels, do you think i should rely on relevant peaks
in amplitude histograms rather than manual observation?

There's no reliable way to measure Po from a multichannel patch. If it's multichannel throughout, try another patch. If parts have only one channel, make a list of those selections and analyze that list.

The manual method:

  • Create a new selection list -- click "AddL/Add List" on the right under Lists
  • Select by hand each burst of single-channel activity and add it to the list -- click "AddS/Add Sel" or press "s" on the keyboard.
  • Set the Data Source (upper-right) to "list" to operate on all your selections
  • Idl, MIL, etc...

The automated method:

  • (baseline must be flat or corrected)
  • Idealize first with a three-color model -- black:closed, red:open, blue:2open
    1. connect black to red, red to blue
    2. select a multi-channel burst, right-click the model, "grab all amps increasing"
    3. Idealize with Data Source: file
  • Find all the single-channel bursts
    1. Determine a cutoff length for the closed intervals between bursts: bt = "Burst Terminator"
    2. Click "Chop Idl" on the right, under Preprocessing
    3. for Burst terminator, choose "Dwell in class 1 longer than bt"
    4. Discard bursts with dwells in class 3 or greater
    5. check "Compute stats"
    6. click OK
  • The Results window should show amplitude histograms for all the single-channel bursts. Po is listed in the Segments tab under "occupancy 2", or you can plot it in the Select tab (Show var -> occupancy 2).
  • The list "Chopped Events" contains all of the single-channel bursts. To operate on all of them, set Data Source to "list".

Custom Histograms 

how can i construct a better amplitude histogram in QuB?
i have seen those in Auerbach and others' papers. i know
that the raw data should be off course fine. but is there
any additional advise from you? by default, QuB seems to
present the Y ordinate as counts/total. but in a paper of
F.Sach (Ding and Sach, BMC Neuroscience, 2002 about P2X2
gating), i found that Y axis has been plotted as counts/bin.
could you tell me how to do that in QuB? i would also like
to know how to calculate the area under curve in QuB?

The most general advice is that QuB can export the table of numbers that make up any histogram. Right-click it and either "Extract" to a tab-delimited text file, or if you have MS Office, "Extract to Excel". Then you can edit those numbers and plot them with your favorite program.

But first have a look at the Histogram window (View menu -> Histogram). The three panels are for amplitude, spectrum, and duration histograms. Click a little icon to generate a histogram from the Data Source (sel, list, file). To customize it, right-click the little icon and choose Properties (make sure to click it again afterwards to re-generate the histogram).

The curves on histograms in the Results window are model-based predictions. They are standard curves (gaussian for amplitude histograms, exponential for duration histograms) whose parameters are output with the histogram -- look in the text output in the Report window, or in the Segments tab of the Results window. For idealization the area is proportional to "occupancy". For MIL the areas are given as "Amp" under Time Constants and in the histograms' tool-tip.

How do you tell how many channels are in your data? 

When we have multiple conductances in the record, how is it possible to
understand if there are multiple conductances of a single channel (for
example 3 or 4 levels), multichannel records or multichannel records with
multiple conductances for a single channel?
Then, can we analyse the multichannel records with QUB? There are options
for N channels in SKM and MIL. If I take in Qub -Preprocessing the
multichannel record, shall I receive the *.ldt file with different levels or
not?

In general you can't tell multiple single channels from a multistate single channel because a whole patch can be viewed as a Markov system as can a single channel. The issue is reallly one of simplicity ofthe model. IF you can explain the data with 4 two-state channels, that is simpler than a single channel of 2^4 states.

What we have done is to first look for bursts that would clearly belong to one object and do the kinetics on that as a minimal model for a single channel. If you have clear bursts, then make a kinetic model of a channel that goes from closed to the bursting mode and see how well a multiple of those would fit the data, compared to a single multistate channel. The best likelihood/paramter criteria (Akaike type criteria) is reasonable to discriminate.

One other simple thing we tried was the following. If you have say 4 conductance levels, closed and 3 equally spaced open conductances you could model that as 3 two- state channels or as a multiconduce mode with 4 states. The independent channel model has fewer parameter (only 2) whereas the serial multistate model has 6 parameters. See what is the difference in the fit. Or, perhaps do 3 identical 3 state models that would have 4 paramters. (paramters=rate constants).

You can analyse multiple or single channels dat wiith QUB. I recommend using MIL since the kinetics of SKM are not reliable. However you can idealsize with SKM and all you need for detection is the simplest serial model with the proper number of visible states. All MIL needs to know is that they are distinguishable, not the kinetics used to idealize the data.

How to choose Dead Time 

I would like to ask what is the best dead time to impose in MIL, and what is
the Sigworth dead time, since it can become quite time consuming to play
with different dead times and then compare fits and LLs.

The sigworth dead time is a good start, but in the presence of noise and the nonlinear properties of the detection algorithm it is not exact. But nothing is since the error depeneds on S/N and rate constants and dt.

A large dead time can fail initially due to a poor starting value plus a big dead time. With td=0, the Q matrix should always have negative and real eigenvalues . But when td>0, this may not be the case anymore. However, if td small (relative to your kinetics), the corrected Q matrix shoud be close to the uncorrected one and have well-defined eigenvalues. That is why we recommend in the paper to start with small dead time and then gradually increase it until reacing a plateau. This will guide the program to avoid the ill-defined Q matrices.

The idea is to pick a dead time, the sigworth value is a good starter, and do a fit. Make the dead time longer (2x?) and try again. The goal really is to find a plateau where the results are not sensitive to the dead time. With some fooling around, you shouldn't have to do a whole screen of every piece of data.

The likelihoods will change as the dead time is changed because that really changes the data being analysed, but the rates should be similar. As with all statistics, you decide how much of a plateau is good. Long dead times, relative the the inverse of the shortest rate constants, are bound to fail, so you probaly will end up in the range of 1-3x the sigworth limit.

The problem is that the dead time correction can lead to complex eigenvalues. There is ontoon implicit in the rates adn dead time, only the ratio matters, so making dt longer makes the dead time correction a smaller factor.

Idealized data from sim has minimal variance since noise has no influence and the sampling clock is linked to the transition clock. That will make it behave somewhat differently from real data.

About MIL Histograms 

1. Are dwell time histograms output after ln or log10 binning of duration intervals.
2. What is the vertical scale is it no/total no of bins
3. When MIL output to _dur.hst file seems to produce a funny graph. It sort
of lookes like the x axis is log binned but the numbers on the x axis are in
ms not ln ms. What is going on there?

The bining is done after taking log10 of the durations. But the x-axis in the plot (and the output hst file) is in absolute time (ms), not log time. In the other words, the spacing of the bins increases geometrically.

The y-axis is scaled by the area of the histogram, so that it can be compared to the pdf. Due to non-uniform binning, the scale is not exactly no/total.

Why is my data broken into segments? 

Hello. If I make a selection that is relatively long (my channel has
clusters of activity that last for several seconds), then when the data is
extracted to an .ldt file or idealized using the SKM method in QUB, the
selection will be broken up into multiple segments. A typical selection
will be 17,000 points, and then it will end up being broken into 3
segments, the largest of which is 8912 points. Is there a way to maintain
the entire selection so that it is not segmented?

A break in segments should indicate a break in the recording. I guess your problem is with the data file format. QuB thinks that pClamp files are segmented, even though they may be continuous. If that's the case, I suggest you extract the whole file with the Join segments option, which will get rid of the segmentation, then do your work on the extracted data.

Problems importing data

QuB can import data in several formats. However, it does not honor all the information in every format. If QuB is misinterpreting your data, try converting it to a different format using the Synaptosoft Mini Analysis Program.

How can I save .dat files with corrected baseline? 

how can i save .dat files whose baseline i have corrected and have that
baseline stay corrected?

You can extract data with corrected baseline. After correcting the baseline,

  • Click "Ext", on the right under "Preprocessing"
  • Under "Data source" choose "Whole file"
  • Under "Pre-process data" choose "As Such" and check "Baseline"
  • Choose an output format
  • Click "OK"

How to plot cluster number vs class occupancy 

with the old version of select i could easily plot out
cluster number vs. class occupancy. However with the update version,
where select has been incorporated into QUB, i can't find a way of
doing this

To plot cluster number vs. class occupancy

  • idealize the data
  • go to the "Results" window, "Select" tab
  • click "Show Var...", and choose "occupancy 2"

If you have MS Office 2000 or newer, you can right-click the plot andchoose "Extract to Excel".

Using the plot to separate clusters

You can group points by color. Change a point's color by clicking it. QUB can make a "selection list" for any color, so you can work with each color separately. Here's how:

(note: I will use the terms point, segment, selection, and cluster interchangeably)

Coloring (grouping) the points

  • click the "Select" button and choose "None". The points become black, which is by convention "not in any group"
  • for each group you see, click an unused color on the left, then click on a few example points
  • to classify the rest of the points according to the example points, click "Select" and choose "K Means"
  • repeat "K Means" until the colors stabilize. If K Means doesn't work, you will have to color all the points yourself.

Working with one group of points (e.g. the red ones)

  • right-click the color red (on the left)
  • choose "Make selection list..." and give it a name
  • go to the data window, find your list under "Lists" on the right, and choose it. The clusters in the list are shown under "Selections".
  • To run MIL on just the red list, set your data source to "List", (halfway down on the right, among "Sel", "List", "File", "File List") and click MIL.
  • To remove the black points from the plot, right-click the color black (top-left of the Select tab) and choose "Remove Segments".
  • To get a new plot with only the blue points, right-click the color blue and choose "Extract Segments"

Some other things to try

  • right-click a point and choose "Show Segment"
  • Move the "SDs" slider (upper right). The horizontal lines are the mean and +/- that number of standard deviations.
  • click "Select" and choose "Trim" to unselect points outside the SD lines
  • click "Show Var" and choose "Custom" to plot e.g. "occupancy 2" vs. "lifetime 2"
  • use the "Criteria" tab to color points using inequalities

Exporting the Results:Segments table 

I am using QuB for the analysis of Maxi-K channel data......After
analyzing the data, I select the results tab and the segments tab......I
cannot figure out how to save or export the data from the table.....I also
cannot figure out why the composite histogram has so few events on the y
axis since each of my 15 segments has over 1000 events.....Thanks in
advance for your help.....

BTW I have been conducting a head-to-head comparison between QuB and
Fetchan for the analysis of patches w/1,2,3 and 4 channels per patch....I
get identical results....QuB is far more user friendly and provides a
wealth of information......Thanks for making it available

The only way to export that table is with copy and paste. We'll look into making a more straightforward option, but for now:

  • Use the mouse to select from the first cell to the last cell
  • Press Ctrl-C (copy)
  • Go to a spreadsheet or notepad
  • Press Ctrl-V (paste)

The histograms are shown as sqrt(count / total). Maybe this is the discrepancy? We may change it to sqrt(count). Which would you prefer?

Thanks for the feedback and compliments.

Trimming baseline for episodic data analysis 

I have only recently began using QUB for data analysis. I'm finding the
software quite easy to use.
My recording protocol includes a baseline before and after the recording
period. Is the only way to exclude the baseline by deletion, or is there a
way to limit the region for analysis.

Yes, there's another way. First, some background:

Analysis buttons such as Idl and MIL can work on the whole file, the selected region, or a list of regions. Some, like MIL, can also work on a list of files. This is the choice on the right of the screen between "sel", "list", "file", and "file list".

The "lists" in question are what you're looking for. They are listed to the right of the data, under "Lists". The contents of the chosen list are shown under "Selections". You can double-click entries to see the corresponding data.

To exclude the baseline regions, build a list containing everything else, and analyze "list". The "Chop" button builds a list by repeatedly skipping N1 data points, adding the next N2 points to the list, then skipping the next N3 points.

To summarize: Chop, choose "list", then analyze.

Trace positioning and vertical scaling 

I am analyzing a data file with two channels....When I open the file and
select a segment (in the Model mode) both the upper and (expanded) lower
windows clearly show the amplitude of the second channel....However when I
apply a baseline correction the amplitude of the second channel exceeds
the height of the expanded window so that I am unable to assign a correct
amplitude for this channel......I cannot figure out how to change the
position of my trace or the scaling of the expanded window so that I may
see both channels.

There's a row of controls along the top which control data display: (auto, +, -, ^, V, ?). They act on the "current" data channel only. To adjust the second channel, set "Ch" to 1 (immediately left of Auto), then use those controls.

Correction of baseline 

I am now trying to correct the baseline in my data. I have a long recording
and the baseline keeps changing as time passes. When I corrected the
baseline with "set as baseline" or "baseline-drift-manual addition of nodes
(highlight data and click "N")", the baseline in all over the data was
changed. As a result, I could correct only small part of the long recording.
How can I correct the baseline in whole data and save it?

The set as baseline only subtracts the meas of the selected region from the data. It doesn't correct drift. Let me suggest that you use the IB function.

Make sure your data is not oversampled. The rise time of events should not be more than one data point. If necessary, filter and resample the data using extract.

Right click on IB to open the options. At the right side where the options are in red, check:

  • Track
  • Correct
  • Back kalman
  • Most likely sequence: gamma

then set:

  • Std 0.01
  • min dy 10
  • FK 10
  • FBK 10

Under idealization options check Baum-Welch

  • Pre process: as displayed

Close the window.

On the top of the page check: Blin apply and show.

Select a chunk of data that contains the relevant jump heights, click selection at the right, and then click IB to correct the baseline on the sample. If it isn't working try varying std in the IB menu. Probably 0.03-0.001 the the maximal useful range of this mysterious parameter (read Lorin's thesis to understand this).

After the selection is working, go back to the IB menu and set:

  • Reestimation: none
  • most like sequence: Viterbi
  • Close window

Then select whole file at the right, and run IB to do the whole file.

IF this doesn't work you can go through the data and insert baseline nodes at good spots in the data and save the corrected data.

Saving idealization with more than one AD channel 

I have a 2-channel data because I am working on gap junctions. When i try to
saved idealized data as a .dwt file, an error message always appears;
"The selected output file format does not allow AD channel count >1. Please
select an apprppriate format."
How can I solve this problem?

Unfortunately we don't have a file format which can save multiple idealized channels. Instead, you can save and re-load each channel's idealization in a separate file:

When you choose File -> Idealized Data -> Save Idealized Data, it saves only the "Active channel" ("Ch", next to "Auto", at the top of the screen). For each idealized channel, make it the active channel then save idealized data.

Alternatively, you can rely on QUB's "session" feature, which preserves idealization, baseline nodes, results, and display settings when you re-open a data file. The data file must be saved in the hard disk or other writable location.

Session information is saved as a .qsf file with the same name and location as the data file. (e.g. C:\data\noname1.ldt and C:\data\noname1.qsf). It is updated automatically each time you close the file. If you move or copy the data file, make sure to move the session file as well.


Running QuB on a Mac

Thanks to jshanata at caltech: 

I just wanted to let you know that I am able to run QuB in Windows XP on
my MacBook, both natively (using Bootcamp to partition my HD) as well as
when I run XP virtually with Parallels.  So far, the only bugs seem to
be with window size/display.


Why don't Idealization stats match MIL time constants?

MIL computes time constants from the optimized rate constants. Idealize computes mean lifetimes by averaging event-duration over all events. They are computed differently, so they shouldn't be exactly the same.

Also, MIL's time constants are modified by the dead time. Prior to QuB 1.4.0.68, Idealize ignores the dead time and reports the mean lifetimes before missed-event correction. In QuB 1.4.0.68 and later, it's an option ("Apply dead time to statistics").


Can I analyze two different channels in a patch? 

I am still doing single channel analysis. It looks fine.
But some data has two channels, one has regular Amp states and the
other has very small Amp states, close to the baseline.
So, I would like to analyze such two channel data with two models.
Could you let me know how I can deal with?

Modeling:mmerge (model merge) builds a model from two or more simpler models. It includes constraints so there are no more free variables than in the source models combined. You can use this model like any other, with one caution: In model properties: rates, check the box "Enforce constraints" so if you edit a rate constant it will update all constrained rates. After MIL, to put the optimized rates back into the source models, right-click the background of the merged model and choose "Un-merge". You might also try removing the "kinetic constraints" from the merged model and re-optimizing, to see if the likelihood improves. If so, the two channels might be interacting.


Help! QuB won't open a data file 

We have a .qdf we can't open...

Here is a general purpose trick for opening any problematic data: first change the file's extension (e.g. ".qdf") to ".dat", then open it in QuB. QuB will prompt you for details of how the data is stored. If you don't know channel count, sampling, scaling you'll have to experiment; closing the file and re-opening until you get them right. QDF files usually have A/D Channel Count: 1, A/D Data size: 2 bytes. Headers and stuff (information about the data, stored in the data file) are mixed in with the data, so have a look through and delete what look like bursts of noise, usually at the beginning or end, or at regular intervals.


How can I discriminate between models with the same LL? 

If I wouldn't know from which starting model I obtained the simulated data file, how could I
discriminate between the two tested models C-O-C (actually the right one) and C-C-O and tell which
model fits data better?

Another question: Why the shapes of the dwell closed times histograms drawn in the "Results" window
and in the "Histogram" window (in both cases "count/total vs Duration(log10) are clearly different?

There are three possible reasons the histograms are different.

  1. Dead time > 0: "Results" histograms use the dead time to remove short events; "Histogram" ones don't.
  2. "Smooth binning" in MIL options: it distributes each event into multiple bins to counteract the effect of sampling; the "Histogram" window doesn't.
  3. "Sqrt ordinate" in MIL options: I think this is to exaggerate the shape of the graph; it's common in the literature, but again, the "Histogram" window does not do this.

The question about how to discriminate between models is more difficult. I think it has been proven, you can't discriminate between two models at equilibrium if they have the same number and color of states. So you need to get it out of equilibrium, or use some other trick.

If the channel binds to a ligand or something like ATP, one of the rates is a binding step that should be proportional to ligand concentration. Try recording the channel at different concentrations: the correct model should be the same at all concentrations, except for one rate that varies proportional to concentration.

Other things you can vary include voltage, pressure, and even point mutations. Also, you can use a stimulus to drive the channel into a particular state, then mark that as the start state in the model. If the models have loops (cycles) you can "Balance all loops" under model: properties: kinetic constraints.

I would recommend looking through our list of Research using QuB to see how other people have done it.


Prev: Multi-channel Popen Outline Next: Reasons to use QuB
Retrieved from "http://www.qub.buffalo.edu/wiki/index.php/Answers"
Personal tools