-
Notifications
You must be signed in to change notification settings - Fork 0
views
This view displays the mean spike waveforms, or spike templates, of each neural unit in the current display focus list on each of up to 16 analog data channels. The templates are laid out in small subplots labeled with the source channel index. As in all the data views, the waveform trace color matches the highlight color assigned to the unit: royal blue for the first unit in the display list -- the primary focus unit --, red for the second, and yellow for the third.
All spike templates are 10 milliseconds in duration, starting 1 ms prior to the spike occurrence time. However, in many cases, the "interesting" part of the spike waveform may only last a few milliseconds. Use the slider control at the bottom left of the view to adjust the visible time span from 10 down to as little as 3 ms. The template span is a user preference that is saved at application exit and restored the next time XSort runs.
Whenever the view is updated, the voltage scale (vertical) is automatically adjusted in accordance with the peak-to-peak amplitude of the primary unit's largest spike template. You can manually adjust the scale between +/-30 and +/-1000uV using the slider control at the bottom right.
NOTE: When there are more than 16 recorded analog channels, XSort only computes and caches each neural unit's per-channel spike templates on the 16 channels [P-8 .. P+7], where P is the index of the unit's primary channel. If two different units do not have the same primary channel, their templates are not computed on the same set of channels. By design, this view displays all 16 templates for the primary unit, as well any templates for other units in the display focus list that were computed on a channel among the primary unit's 16 template channel indices. This makes sense -- it's unlikely the user will need to compare units with very different template channel sets.
This view displays the autocorrelogram (ACG) of the spike train for each unit in the current focus list, as well as the crosscorrelogram (CCG) of each unit's spike train with the spike train of a different unit in the list. If only one unit is selected for display, only its autocorrelogram is shown. If two (or three) units are selected, the correlograms appear in a 2x2 (or 3x3) array of subplots, with the autocorrelograms on the main diagonal. Note that the CCG trace colors are a blend of the colors assigned to the units in the display list (red + blue = magenta, etc).
The correlograms are computed once in a background task, then cached in memory until the working directory is changed or the application exits.
As with the Templates view, you can "zoom" in on the correlograms; use the slider at the bottom of the view to adjust the correlogram span between 20 and 200 milliseconds (+/-10ms to +/-100ms).
The plots include two annoations: a dashed white line at zero correlation and a translucent white vertical band spanning T=-1.5 to +1.5 milliseconds. These visual guides help the user quickly check whether or not there is significant correlation in pairs of spikes occurring 1ms apart -- an indication that the neural unit's spike train is contaminated in some way. The "Show zero correlation" checkbox toggles the visibility of the zero correlation line.
Both the visible state of the zero correlation marker and the current correlogram span are user preferences that are saved at application exit and restored the next time XSort launches.
This view plots, on the same axes, the interspike interval (ISI) histogram for each unit in the display focus list. The histogram is an array of normalized bin counts for ISIs between 0 and 200 milliseconds, with each bin count divided by the maximum observed bin count. As with the correlograms, the ISI histograms are computed once on a background task, but then cached in-memory so they need not be computed again.
Use the slider at the bottom of the view to adjust the visible histogram span between 20 and 200 milliseconds.
This view renders, for each unit in the display focus list, a 3D histogram representing the unit's autocorrelogram as a function of instantaneous firing rate. This can be thought of as a "3D autocorrelogram" that shows firing regularity when the unit is firing at different rates. The histograms are rendered as 10x201 heatmaps, with observed firing rate divided into 10 equal bins along the vertical axis, and time T relative to spike occurrence along the horizontal axis, with T between -100 and +100 ms. The number of firing rate bins and the correlogram span are fixed and cannot be changed. The subplot title displays the unit UID and the observed range of the instantaneous firing rate for that unit.
Like the ACGs, CCGs, and ISI histograms, the ACG-vs-firing rate histogram is computed once in the background and then cached in memory.
Use the slider at the bottom of the view to adjust the visibile histogram span between 20 and 200ms (+/-10ms to +/-100ms).
This view displays, for each unit in the current display list, a histogram of firing rate as a function of time over the duration of the recording session. It provides a means of quickly checking whether firing rate changed dramatically at any point during the recording (eg, if the unit was "lost" at some point during the experiment).
The firing rate histograms are drawn on a time scale spanning the entire recording, which may last an hour or more. Use the combo box at the bottom of the view to select a bin size between 20 and 300 seconds. If the Normalized box is unchecked, the Y-axis is firing rate in Hz. If checked, the histograms are normalized by dividing each bin by the unit's mean firing rate.
As you move the cursor across this view, a white vertical line follows the cursor, and a text label near the top of the line displays the elapsed time of the recording at that point in minutes and seconds -- MM:SS. A dot marks the intersection of that line with the firing rate histogram, and an accompanying label displays the normalized ("1.08") or unnormalized ("24.3 Hz") firing rate for the corresponding bin.
A dashed green vertical line indicates the elapsed recording time T at which the analog data clips start in the Channels view. If you change the start time T in the Channels view, this green line is updated accordingly. Conversely, you can change the clip start time by clicking anywhere inside this view. This is helpful if the firing rate histograms unveil some sort of anomaly and you wish to see what's happening at that point in time in the analog data recording.
This view renders the results of a principal component analysis (PCA) on the spike clips of each neural unit in the current focus list. The purpose of the analysis is to "map" each spike in each unit's spike train to a point in a two-dimensional space, offering a "scatter plot" to help assess whether the units are truly distinct from each other. The PCA view renders each unit's scatter plot in the assigned color (blue, red, yellow), provides for downsampling of the scatter plots when the unit(s) contains hundreds of thousands of points, and allows the user to define a split region in PCA space (via a sequence of mouse clicks inside the view) when only a single unit occupies the display list.
By design XSort only computes mean spike waveforms -- aka spike templates -- on a maximum of 16 channels "near" the unit's primary channel; call this the unit's template channel set. If N<=16 analog channels were recorded, then all units will have the same template channel sets. PCA is restricted to a unit's template channel set to keep the computation time and memory usage reasonable. This is very important for recording sessions with hundreds of analog channels.
When a single unit is selected for PCA, only the channels in that unit's template channel set are included in the analysis. When 2 or 3 units are selected for PCA, only those channels comprising the intersection of the units' template channel sets are considered. If the intersection is empty, then PCA cannot be performed.
Let N = N1 + N2 + N3 represent the total number of spikes recorded across all K units (we're assuming K=3 here). Let the spike clip size be M analog samples long and the number of analog channels included in the analysis be P. Then every spike may be represented by a multi-clip: a vector of length L=MxP, the concatenation of the clips for that spike across the P channels. The goal of PCA analysis is to reduce this L-dimensional space down to 2, which can be easily visualized as a 2D scatter plot.
The first step is to compute the principal components for the N samples in L-dimensional space. A great many of these clips will be mostly noise -- since, for every spike, we include the clip from each of up to 16 channels, not just the primary channel for a given unit. So, instead of using a random sampling of individual clips, we use the spike templates computed on each channel for each unit. The per-channel spike templates are concatenated to form a KxL matrix, and principal component analysis yields an Lx2 matrix in which the two columns represent the first 2 principal components of the data with the greatest variance and therefore most information. However, if only 1 unit is included in the analysis, we revert to using a random sampling of individual clips (because we need at least two samples of the L-dimensional space in order to compute the covariance matrix).
Then, to compute the PCA projection of unit 1 onto the 2D space defined by these two PCs, we form the N1xL matrix representing ALL the individual spike multi-clips for that unit, then multiply that by the Lx2 PCA matrix to yield the N1x2 projection. Similarly for the other units.
[NOTE: From this description of the PCA algorithm used, it should be clear that, whenever the composition of the display focus list changes, the principal component analysis must be redone from scratch.]
PCA is a time-consuming operation that is always performed on a background thread. It can take many seconds to complete, especially if the unit spike trains are very long, and can impact the responsiveness of the user interface (whenever you change the composition of the display list, any statistical computations going on in the background must be cancelled before the views are updated). XSort will not queue a PCA computation at all if the PCA view is hidden -- so it is best to hide the view until you actually want to see the PCA projections.
When there are a great many points to draw, it takes a noticeable amount of time to render the PCA projection scatter plots . And with a great many points and overlapping projections, one unit's projection can obscure another's. At the bottom of the view are two controls to help address these issues:
- Pressing the Toggle Z Order button changes the Z-order of the displayed scatter plots when more than one unit is in the display list.
- The Downsample combo box lets the user choose the downsampling factor for the rendered plots. The higher the value, the faster the scatter plots will render -- but only because fewer points are drawn. Set this to 1 to disable downsampling.
The main purpose of the PCA view is to identify a neural unit that is really a collection of two populations of spikes belonging to two distinct neurons. For such a unit, the PCA scatter plot might consist of two physically distinct "blobs". When a single unit occupies the display list and its PCA projection has been fully calculated, you can use the mouse to define a closed polygonal region around one of the blobs -- the split region. Simply click inside the view to start the polygon path; once you relese the mouse button and move it (don't hold and drag), a thin white line will follow the cursor as it moves within the view. Click again to define a second point in the polygon, and repeat this process until you've surrounded the "blob" you wish to isolate. Double-click to define the last point and close the polygon region. You can always start over if you've missed some points. When satisfied, choose the Edit | Split menu command to split the neural unit into two derived units -- one containing the collection of spikes that project inside the polygonal split region, and the other formed from the spikes that lie outside it.
See the Making Changes section of the user guide for additional details about splitting a unit, as well as merging two units, deleting a unit, or undoing those changes.
This view displays a one-second clip (or shorter) for each of up to 16 analog data channels. When a working directory is first opened, traces for the first 16 channels (0-15) are shown. Once a primary focus unit (royal blue) is selected in the neural units table, then the channels displayed are those from the unit's template channel set (channels [P-8 .. P+7], where P is the unit's primary channel index). In addition, for each neural unit in the current display focus list, the view superimposes -- on the trace for that unit's primary channel -- 10-millisecond clips indicating the occurrence of spikes from that unit. In keeping with the other views, these spike clips are rendered in the highlight color (blue, red or yellow) assigned to that unit.
A slider at the bottom of the view lets the user choose any 1-second segment over the entire analog data recording. The companion readouts reflect the elapsed recording time -- in the format MM:SS.mmm (minutes, seconds, milliseconds) -- at the start and end of the currently visible portion of the traces.
With the mouse cursor inside the view, use the mouse's scroll wheel to zoom in or out on the plotted traces both horizontally in time and vertically in voltage, then hold down the mouse and drag to pan the view in either direction. The plot's x- and y-axis range limits are configured so the user can zoom in on any 100ms portion of the 1-second segment, and on any 2 adjacent channel traces. To reset the view, click the small icon ("A") located in the lower-left corner of the view.
Each time the channel traces are updated, XSort adjusts the vertical scale in accordance with the worst-case peak-to-peak voltage excursion observed across all the traces. You can manually adjust the scale between +/-50 and +/-1000uV using the slider control at the bottom right.