Using the HDCV Analysis Program

Transcription

1 Using the HDCV Analysis Program This manual describes Analysis.exe, the data analysis portion of the HDCV (High Definition Cyclic Voltammetry) program suite from the University of North Carolina. Contents Time and Terminology.. pg. 1 Data Files pg. 1-3 Opening a File. pg. 2-3 Crash Recovery... pg. 3 The Main Display.. pg The Color Plot.. pg. 4 The I vs T and CV Graphs. pg. 4 The Digital State Graph.. pg. 4 Navigating through Time pg. 5 Navigating Long Files. pg. 5-6 Options on Load.. pg. 6-7 Channels and Waveforms.. pg. 7-8 Digital Background Subtraction. pg. 8 Smoothing pg Exporting What You See pg Cleaning and Organizing the Data.. pg Snippets pg. 12 Navigating Snippets pg. 13 Manual Snippet Creation pg. 13 Glitches. pg Saving Your Work pg. 14 Concentration Analysis... pg Quick Concentration Conversion.. Chemometric Analysis General Theory of PCR.. The Training Set.. Sanity Checking... The Concentrations Screen... Printing and Exporting Your Data. pg. 15 pg. 15 pg pg pg pg pg. 20

2 Data Averaging.. pg Averaging Files pg. 21 Averaging Runs.. pg. 21 Averaging Snippets. pg The Averages Screen. pg. 22

3 Time and Terminology Before plunging in to the details of the program, we'll take a moment to establish some terminology about units of time. The smallest unit of time in HDCV is the sample time. This is the interval at which each data point in the waveform is generated and measured, typically on the order of 10 microseconds. All other times used in HDCV are multiples of this one. Its inverse, the sampling rate, is a commonly quoted figure. Next is the interval at which the HDCV ramp is output. This is the frame time, typically 100 milliseconds. The time from the beginning of one ramp to the beginning of the next is one frame. Its inverse, the cyclic voltammetry frequency or CVF, is also commonly mentioned. The cyclic voltammetry scan only occupies a subset of the frame time. This is the scan, and its duration is the scan time. In the case that multiple different waveforms are being output on different channels, the frame begins when the first waveform begins, and the scan time runs until the last waveform ends. The remainder of the frame is the rest time. A common type of CV experiment consists of a series of timed events in which some stimulus is given and CV data are recorded for an interval before and after that time. Each such event, with the recording of data, is one run. The entire series of runs, the largest unit of operation that HDCV will automatically control, is the experiment. There are also experiments that are run in continuous mode, where data is collected continuously throughout, and not broken up into runs. 1. Data Files The HDCV Analysis program is designed to read files from the older TarHeelCV, and from HDCV itself. These are the file types it recognizes: A. TarHeelCV data file. These typically represent a short period of time, up to about a minute in length. Unfortunately, they have traditionally been stored with no filename extension, which means that the open file dialog has to accept and display files of all extensions. Let the user beware: trying to open a file that is not a cyclic voltammetry data file will produce unspecified amounts of unhelpful results. B. HDCV files converted from groups of TarHeelCV data files. Such groups of files are usually named with consecutive numbers: 100, 101, 102, etc. The HDCV Data Converter program will convert and concatenate a group of such files into a single, 1

4 longer HDCV format file. TarHeelCV data files are often accompanied by digital data files consisting of the same file name with the.txt extension. These are text files giving the state of digital input lines at each frame. When expect digital data is checked in Data Converter the program will also output a.dig file along with the new.hdcv file. HDCV Analysis will display the timestamps saved in the.dig file when opening the main.hdcv file, as long as they both have the same file name and are located in the same directory. C. Data files originally recorded with HDCV. These are recorded in one of three modes: - Single. The file records a single, automatically timed event (stimulation, drug injection, etc.). - Multiple. The file records a series of runs, each of which represents one automatically timed event. Often several minutes of unrecorded time elapse between events. - Continuous. The file records data collected continuously over a longer period of time, which may be multiple hours. This is usually used in conjunction with behaving-animal experiments. HDCV format files have the.hdcv extension. In some cases, especially with behavinganimal experiments, they are associated with digital data, stored in a file with the.dig extension. These.dig files, unlike the older.txt data files, record only transitions in the digital state. They can easily be viewed with Notepad. Opening a File One may specify a file to be opened by clicking the folder icon beside the file path box at the upper left of the screen, by entering an absolute pathname directly in the file path box and pressing Enter, or by clicking the Load button when the file path box is empty. Clicking the Load button when the file path box is not empty will reload the file named therein. Upon successful loading, the color plot graph in the upper left region of the window will fill with color. Several attributes of the file are displayed. The date and time at which recording began are displayed in the Date Stamp box, centered about an inch below the color plot. In the case of a Continuous file, the duration of the file, in seconds, will be displayed beside the date stamp. (In the case of a Single file, the file time can be read from the time scale of the color plot; it displays the entire experiment. In the case of a Multiple file, file time is not a very meaningful statistic, since there are gaps.) In the center-right area of the window, the file Header is displayed in brief. The file header is 2

5 text containing information about all the parameters of the experiment. It is not always pretty, and it sometimes says the same thing more than once, but it contains enough information to reconstruct the experimental parameters if need be. Clicking the Show Full Header button will bring up a window in which one may read the entire header. Crash Recovery If the data file appears improperly closed due to an abrupt end to the experiment (the computer lost power, Windows crashed, etc.) you will be asked for permission to repair the file. If the file statistics it presents to you seem reasonable, say yes. This is the only case in which Analysis will ever make any change to your data file. 3

6 2. The Main Display The upper portion of the display contains four graphs: the Color Plot, Current Versus Time (I vs T), The Cyclic Voltammogram (CV), and the Digital State graph (below I vs T). The Color Plot The color plot presents a two-dimensional array of data. Each column represents one scan, the time of one CV ramp. Adjacent columns represent consecutive scans. Color represents the current measured at each data point of each scan. The I vs T and CV Graphs The primary cursor (white) in the color plot graph selects one row and one column for further display. One row of the color plot represents a given data point selected out of each scan. This is precisely the information that is displayed in the I vs T graph, for the cursor row. Height in this graph corresponds to color in the color plot. The cursor column selects one scan. This information is displayed in the CV graph on the right. Actually, that isn't the whole truth. Based on user options, I vs T may actually display the average of a few consecutive data points near the cursor row, and CV may display the average of a few consecutive scans near the cursor column. See Options on Load. This helps cancel out noise and produce smoother plots. Move the cursor to see the other graphs updated in real-time to reflect its new position. Sometimes the displays may get behind while the cursor is in motion, but they should catch up quickly once it stops. You can also see, below the color plot, numeric readouts of the selected data point and scan time, and the current at that time. Edit the data point or time fields to numerically position the cursor. If a stimulus train was applied during the timeframe of the file, two blue indicator boxes will be shown on the I v T trace. These boxes mark the onset and termination of the stimulation. A more detailed record of the stimulation parameters is viewable in the Show Full Header dialog box. The Digital State Graph The Digital State graph displays data only when digital data (.dig) were found during file loading. It displays the state of each digital line. If these lines were given names when the file was recorded, those names will appear in the legend along with the colors. 4

7 Navigating through Time Below the color plot you will see dials named Pan and Zoom. These names are taken from video technology: Zoom will zoom in on detail or zoom out to a more inclusive view; Pan will move laterally through time, while maintaining the same scale. Rotate the Zoom dial to the left for maximum detail; rotate to the right for maximum overview. Analysis will never zoom out to display more than five minutes of data at one time, because this tends to cause LabVIEW to run out of memory. When displaying a Multiple mode file, Analysis will display navigation buttons in a second row below Pan, Zoom, Data point, and Time. The "go to start" button will go to the first run in the file. The "backward" button will go to the run previous to the one currently displayed, unless the first run is already displayed. The "forward" button will go to the next run. The "go to end" button will go to the last run in the experiment. Navigating Long Files Navigation becomes more interesting in the case of long, Continuous mode files. When displaying these files, Analysis cannot load all the data at once. You may notice this because certain navigation operations take longer than others. Here is how it works: the "view range" is precisely the range of time currently visible in the color plot. The "load range" is the range of time currently loaded into memory. If any navigation operation creates a new view range that is not a subset of the load range, the old loaded data will be cleared from memory and a new range will be loaded, precisely equal to the currently needed view. 5

8 Continuous mode files also use the navigation buttons, but in a somewhat different way from Multiple mode files. All four buttons will operate at whatever time scale is currently set by the Zoom dial. Let us refer to the amount of time that will fit in the color plot at the current zoom scale, as one "span". The "go to start" button displays one span starting at the beginning of the file. "Backward" displays the span immediately prior to the one currently displayed. "Forward" displays the span immediately after the one currently displayed. "Go to end" displays the last span, up to the end of the file. Along with these methods, one can also directly edit the limits of the scale, to view a precisely chosen range. Options on Load The following options take effect every time a file or a portion of a Continuous file is loaded: Filtering on. If this is on, a low pass filter is applied to the data as they are read, to reduce high-frequency noise. The default is a fourth order Bessel low pass filter with a cut off of 2 khz. Click the "Filtering..." button to see and change these options. Besides Bessel, several other filter types are provided, whose mathematical details are beyond the scope of this manual. Most people just use the default Bessel. Auto find max. If this is on, as soon as the color plot is displayed, the cursor automatically goes to the point with the highest current reading. Hopefully this will turn out to be the peak of interest. This option does not apply to Continuous files. Auto background subtract. If this is on, the data are background subtracted as they are loaded. The background is taken, by default, as the first 10 scans, averaged. This is usually good enough to get a reasonable picture. See Background Subtraction, below, for ways to refine this choice. Averaging settings. The user may change the number of CV s and I v T s that are averaged around the cursors in the color plot by opening the Options dialog box. These should be set appropriately for the collection parameters used (see Wiedemann et al, Anal Chem, 1991, 63, ). The default settings are configured for the catecholamine waveform (-0.4 V to 1.3 V triangular ramp, 400 V/s, 850 data pts, 10 Hz application) and are as follows 6

9 o Rows to average for I vs T: 5 o Columns to average for CV: 5 o Columns to average for background: 10 Remove points from start of scan. This feature, also available in the Options dialog box, allows the user to remove a given number data points from the beginning of the CV scan. It is most commonly used when analyzing data that has been collected with analog background subtraction (see the HDCV Data Acquisition) or when analyzing combined electrophysiology/electrochemistry files collected with TarheelCV. In combined the combined ephys/echem mode, the potential of the electrode floats between voltage ramps to record cell firing activity. When a voltage is again applied to the electrode a current spike occurs at the beginning of the scan. To prevent this artifact from occurring during the voltage ramp, buffering points are added to the beginning of the scan, which are then subtracted during analysis. HDCV Data Collection executes this combined mode by not recording during the buffering period, and therefore does not require data point removal for analysis. Set time zero. By default HDCV displays time = 0 as the beginning of the file. It is sometimes more appropriate however to set time zero to an output event such as stimulation or digital TTL. The timing of the displays can be reconfigured by accessing the Options dialog box. This will also transfer to the color plot and I v T text files when exporting the data. Channels and Waveforms HDCV supports the use of multiple electrodes, possibly driven by different waveforms. There may be up to 4 different waveforms, distributed among up to 16 electrodes. Each electrode is referred to as one channel. When setting up the experiment, the experimenter has the option of naming the channels and the waveforms. Otherwise, they will be numbered. Below the Pan dial you will see the Chan(nel) control. Change its value to see data displayed for a different channel. Below the CV graph you will see a Waveform menu. HDCV records the relationship of channels to output waveforms, so hopefully you should never have to use this menu except as a comforting confirmation. But if the relationship was not specified correctly at experiment setup time, you may override it here and select the waveform to use for viewing the cyclic voltammogram. Warning: choosing the wrong waveform will 7

10 probably produce a CV that is utter gibberish. Digital Background Subtraction Press the button with a green vertical line to summon the green Background cursor. It appears, if it wasn't already there, at the left edge of the color plot. Drag the background cursor to choose the time that will be considered as background. Whenever you release it, background subtraction is adjusted to treat that time as the background. Caution: if you leave the background cursor on top of a peak or other interesting event, that moment in time will be flattened and all others will look very strange indeed. Smoothing Smoothing is another level of noise reduction filtering, in addition to the Bessel filter described above. It is applied two-dimensionally to the color plot data. Where the Bessel filter will only influence the smoothness of each individual column in the color plot (in the voltage domain of a single CV), Smoothing will also smooth out differences between nearby columns (both in the voltage domain within a single CV and among consecutive CV scans). It is, in effect, a two-dimensional low pass filter. There are two methods of smoothing provided in HDCV: a convolution kernel and a 2D fast fourier transform filter. The Smooth button will apply whichever method has been configured by the user most recently. The convolution kernel has been retained for compatibility with TarHeelCV. Click the "Kernel..." button to see its options. You may use any size of kernel from 3x3 (the smallest meaningful size) up to 9x9. It may have different sizes in the two dimensions. There are three approaches: - Gaussian. A Gaussian kernel has many nice properties, including circular symmetry and a separable (more efficient) nature. Usually this is the best choice. - Separable. A separable kernel has the property that any row is a multiple of any other row, and any column is a multiple of any other column. Thus if one row and one column are defined, the entire kernel is known. This allows an especially efficient implementation: the time depends on the kernel size as N + M rather than N x M. - Fully custom. You may choose every element of the kernel. It is quite possible to make noise enhancing, or edge enhancing, kernels as well as noise reducing kernels. To reproduce the 1 pass smooth from TarheeCV, use a 3x3 Gaussian kernel of 1 s. 8

11 The newer method is to operate on a two-dimensional Fourier transform. A twodimensional Fourier transform converts the color plot from the image domain to the frequency domain, and allows frequency domain filters to be applied. The figure below illustrates this process: A 2D FFT converts the color plot shown in the Pre FFT window to the frequency spectrum shown in the Transformed view window. In Transformed view, the x- axis represents frequencies along the time axis of the color plot, while the y-axis represents frequencies along the CV axis of the color plot. The origin (0 Hz) of Transformed view is at the center of the window, and frequencies extend from fs/2 to fs/2, where fs is the sampling frequency. The sampling frequency in the data point domain is the sampling rate and in the time domain it is the cyclic voltammogram frequency. In this example, the color plot time sampling frequency is 10 Hz, so the Transformed view x-axis extends from -5 Hz to 5 Hz. The color plot CV sampling frequency is khz, so the Transformed view y-axis extends from khz to khz. Filtering is accomplished by drawing an ellipse in the Transformed view. Frequency content within the ellipse is preserved, while content outside the ellipse is discarded. 9

12 Drawing an ellipse at the center of the window (centered on 0 Hz) is a low-pass filter, and the cutoff frequencies of the ellipse are shown in the Time cutoff (Hz) and DP cutoff (KHz) fields. By selecting different cutoff frequencies in the x and y directions, differing amounts of filtering can be applied in the time and CV directions of the color plot. To prevent ringing artifact, the transition at the edge of the ellipse must be smooth. The Rolloff style dropdown box allows selection of the transition function. In this case, a Bessel function is used. The two ellipses shown in the Transformed view illustrate the width of this transition. The inner ellipse (bright green) is the -3 db point (1/2 power cutoff) of the Bessel function, and the outer ellipse (dark green) is the -50 db point (full power cutoff) of the same Bessel function. The numbers reported in the Time cutoff (Hz) and DP cutoff (KHz) fields correspond to the -3 db points of the function. For the Bessel function, the -3 db point is calculated by multiplying the sampling frequency by For the -50 db point, multiply the sampling frequency by To replicate the 3x3 smoothing kernel implemented in Tarheel CV, use a Bessel function with cutoffs of 0.135*fs in HDCV. After the frequency domain filter is applied, the resulting spectrum is shown in the Clipped Transform window. In this example, the frequency content between 0 Hz and 3.35 Hz (-50 db point) along the time axis is preserved, and frequency content between 0 Hz and khz (-50 db point) along the CV axis is preserved *. An inverse 2D FFT converts this modified frequency spectrum back into an image, which is shown in the Post FFT window. The effects of the frequency domain filter are apparent in that the Post FFT color plot looks smoother, or has less high frequency content, than the Pre FFT color plot. The filtered I vs. T and CV plots are also shown on the right side of the FFT Setup window. Frequency domain filtering has advantages over conventional techniques, including more flexibility in establishing cutoff frequencies, and no peak shifting in the filtered plots. Exporting What You See Wherever you see the word "export" in this program, it refers to writing data out into text files. In particular, it writes the tab-delimited text format, which is a universal standard. All common spreadsheets, word processors, and databases, as well as Matlab and many other programs, can handle this format gracefully. 10

13 The Export button on the main display generates three such files, from the three graphs as currently viewed: filename Color.txt The color plot, as a large two-dimensional layout of numbers corresponding to the rows and columns of the color plot. Each number represents the current as measured at that point. filename datapoint IT.txt The I vs T plot at the data point marked by the white cursor, in two columns: time on the left, current on the right. filename time in file CV.txt The CV plot at the time marked by the white cursor, in two columns: data point on the left, current on the right. Notice that if you move the white cursor around and click Export again, you will get I vs T and CV files that are differently named, but you will get a repeat of the color plot file, and you will be warned and asked if you want to overwrite it. The answer is that it doesn't matter, as long as you haven't panned or zoomed or otherwise change the view between times. But if you would like to create more distinction among the files you save, put text into the "suffix" field, and it will be added to the generated file name right after the name of the data file you are working on. 11

14 3. Cleaning and Organizing the Data The lower portion of the Analysis window contains two tabs. Click on the Snippet and Glitch to enter this section. Snippets Continuous files, in reality, usually consist of a series of events of interest, often with expanses of uninteresting time in between. Frequently these events of interest coincide with digital input signals. We use a MedAssociates box with an 8-bit digital input interface to the computer. An event of interest such as a cue, lever out or lever press will usually be marked by signals on this interface. The analysis of these files begins by forming "snippets". A snippet is defined as a specified number of seconds before and after one of these digital events. Or sometimes, there is a two-event snippet: a specified number of seconds before a first event through a different, second event and a specified number of seconds following. Typically each snippet contains one dopamine peak or other chemical signature event. Press the "Find snippets..." button to specify snippets. Here you have the opportunity to specify the digital input line or lines that mark the snippet, and whether to time the snippet based on a rising or falling signal. Below you can specify the length of the snippet before and after its defining signals. Ideally the length will be long enough to include all data of interest surrounding the event, and short enough not to include peaks from any other events. The "Max timespan" field is useful when defining twoevent snippets, to prevent any excessively long snippets from being formed. Press the OK button to proceed. Now you will want to use the "Snippet Navigation" section. Four buttons allow navigation to the first, previous, next, and last snippets, respectively. The number of fields will show the total number of snippets and which one is showing. You can type in the number field to go directly to a numbered snippet. This is a good time to make a first examination of the data. Some snippets may prove to contain anomalous, jarring events, typically caused by the animal making an abrupt movement. If you see one of these, press the Exclude button. You will not have to see it any more, nor will it be included in any averages. If you will be using chemometric analysis, you will get another chance to exclude bad snippets later. Some snippets may prove to be particularly interesting for some reason. You can press the Flag button to mark such an event for future reference. 12

15 Navigating Snippets Use the four navigational buttons to go to the first, previous, next, and last snippet in the list, respectively. Normally once you exclude a snippet, you will never see it again. But if you did this by mistake, you can use the green/red previous and next buttons, and these will take you sequentially through all the snippets including excluded ones. You can then click Exclude again to accept it. Likewise, when you see a flagged item, you can click Flag to unflag it. Manual Snippet Creation Sometimes snippets may be useful even in a data file without digital signals. In this case you can create them manually. Simply navigate to view the range that you want to use as a snippet, and then press the Manual Capture button. A snippet is created that consists of precisely the viewed range. Manual snippets can only be created from HDCV files collected in continuous mode and from TarheelCV files processed by the Data Convertor VI. Captured manual snippets must be of the same length to be averaged in the simple concentration or chemometrized concentration screens. Glitches Sometimes while viewing the color plot you may see a tiny, colored region in a place that bears no relation to a meaningful event. Often it will occupy only one piece of one column of the color plot. These are glitches, little bits of electrical interference unrelated to our experimental data. To the intelligent viewer of a color plot, they are easily ignored. But they can do a lot of damage to some of our statistical methods for computing concentrations. Hence it is a good idea to deal with them now. This is the issue of glitch repair. To repair a glitch, move the white color plot cursor to point to the glitch -- or preferably, one or two columns to either side of it. Be sure that the cursor is vertically aligned with the glitch, as this will make it easy to see the glitch on the I vs T plot. then press the (green stripe on black background) button to bring the green glitch cursor on screen, if it is not visible already. It appears in the I vs T plot. Move the white cursor to one side of the glitch, and the green cursor to the other, and closing it as tightly as possible. Then press Fix Glitch. The glitch vanishes, replaced by a flat continuation of the data values just before the glitch. Exception to this procedure: if the glitch falls in the middle of a meaningful event, the most appropriate response is generally to press the Exclude button, removing this event 13

16 from any further consideration. Saving Your Work At the bottom of the snippets tab you will find three buttons which allow you to save and recall analysis settings for a given file. Pressing the Save Work button will create a companion file with the same name as the original HDCV file but with a.aly extension. This.aly file records filtering parameters, Options settings, cursor placements, deglitching information and snippet timing. As long as the.aly file is in the same directory, the program will automatically load any saved analysis settings when opening the.hdcv file. You can also save multiple.aly files for a single data set by selecting the Save As button. If you wish to apply settings from a.aly file with a different base filename or directory, use the Load button. 14

17 4. Concentration Analysis Quick Concentration Conversion A current versus time trace can be converted into a concentration versus time trace using HDCV s quick concentration convertor. After selecting a current versus time trace by moving the cursor on the color plot, input a calibration factor in the cal factor box above the I vs T display on the main screen. This factor should be in units of concentration/ na signal. Once an I vs T and calibration factor have been designated, select the Quick Conc button underneath the CV display. This will prompt a new screen showing the original color plot and the converted concentration versus time trace. Controls below the concentration display allow you to toggle between multiple runs and snippets. The quick concentration feature should only be used when the current trace represents a single analyte. In vivo this is not always the case. When recording in the brain of an awake animal for instance, electrically-stimulated dopamine is often accompanied by pronounced ph changes. These ph shifts also produce current at the oxidation potential for dopamine. To handle these cases we call upon a multivariate analysis technique called principal component regression. The following sections describe this in detail. Chemometric Analysis Chemometric analysis, or Principal Component Regression (PCR), is a statistical method that allows us to remove noise from our data and estimate concentrations of one or more substances in mixed and murky situations, given a few data samples that are trusted to be clear and pure. These trusted samples are known as "training standards", and collectively as a "training set". General Theory of PCR A plot of peak current versus concentration is termed a univariate calibration curve because intensity is measured at only one potential. While this approach is simple and fast, it is incredibly limiting. Why collect entire CVs if all we ever use is one potential value to ascertain concentration information? Moreover, multiple species (and noise) can contribute current at a specific potential so, in a situation where multiple species are signaling simultaneously in vivo, we may be unable to parse out how much each distinct 15

18 substance is changing. We scan multiple potentials because most substances have uniquely shaped cyclic voltammograms and their shapes can be used to confirm identities and distinguish between different species. Current isn t just proportional to concentration at a specific potential; current is proportional to concentration at all potentials of a voltammetric scan. Therefore, we could make a calibration curve where we plot current at multiple potentials versus concentration. Such a calibration curve is termed a multivariate calibration curve because current at multiple potentials is used for concentration prediction. PCR is one method of multivariate calibration where current values at all of the points on a CV are related to concentration. When you think of a calibration curve in terms of an electrochemical measurement, you may think of plotting peak current versus concentration. To construct such a univariate (one potential) calibration curve in vitro, you would make a set of standard solutions of a specific species at known concentrations. You then train a computer to learn the relationship between intensity and concentration. It uses regression analysis to deduce a linear relationship between intensity and concentration (Y = MX + B). Finally, you use this linear relationship to predict concentration values in an unknown measurement. PCR works in the same way, but it uses current values at all potentials to construct the calibration curve. Just like for a univariate calibration curve, a set of training standards are needed so the PCR algorithm can learn how current and concentration are linearly related. Then, this relationship is used to predict concentration values for an unknown. For more information, see Keithley, Heien, and Wightman, TRAC-Trends in Analytical Chemistry, 2009, 28, and the correction in Keithley, Heien and Wightman, TRAC-Trends in Analytical Chemistry, 2010, 29, 110. The Training Set If you will be using the chemometric features of this Analysis program, click the Training Set tab in the lower portion of the main Analysis window. Here you will instruct the program on what substances to look for and how to look for them. The leftmost column is titled Analytes. List here the things you will be looking for. Most commonly we use two: Dopamine, and ph. Beside each analyte name provide the scaling factor. This factor converts a current reading into concentration units appropriate to the analyte. Scaling factors may sometimes be negative, for analytes that have a negative impact on CV current. For each analyte you will now select training standards. In the "Working on" column, 16

19 click the button beside the analyte you want to train first. Now you will go and find training standards. Probably in the early portion of the experiment you have created training files while adjusting electrode depth. Open one of these (using the filename field at the top of the window) and look around until you find a clear, clean dopamine CV (or CV suited to whatever analyte you are working with). Press Capture Standard. It will be added to the list of standards, and a trace will appear in the Training Set graph to the right. Repeat this process until you have captured enough standards (at least five). The standards you capture should span the range of concentrations expected in the data to be analyzed. The more successful you have been at capturing good, standards, the more nearly the traces in the Training Set graph will be linear multiples of each other. If you capture standard and then don't like it, uncheck the checkbox beside it in the standards list; that slot will be cleared or reused at the next opportunity. But if you change your mind before capturing any more standards for switching analyte, you can check the checkbox back on to keep it. The above process must be repeated until you have at least four standards for each analyte. Input the calibration factor for each analyte in the column titled factor. The calibration factor should be in units of concentration/ na signal, and should be based on flow cell responses measured for a given electrode type and size. After this is completed press the Train button. This uses the training standards to compute matrices that will be used for calculating concentrations hereafter. The quality of any calibration curve depends on the quality of the training standards ( Garbage In, Garbage Out ). In addition, the closer the training standards are to the unknown measurements you wish to predict, the more representative,applicable, and accurate the calibration curve is for predicting concentrations in your unknown. Here are some general tips for generating training set standards (taken from Keithley, Carelli, Wightman, Anal. Chem., 2010, 82, and Keithley and Wightman, ACS Chem. Neurosci., 2011, 2, ). 1. Training sets must contain CVs of all expected components in the unknown measurement. 2. The noise in your CVs is used to calculate the residual threshold (see below) so the noise level in training set CVs must be representative of the ones in your unknown measurement. Therefore, do not bias your training set standards to try include less noise. This will lower the residual threshold to an artificial value lower than necessary causing you to discard data you could normally use. Alternatively, do not bias your training set to include standards with high noise values. This will raise the residual threshold to an artificially high value that would cause you to keep bad data. 3. Do not to take more than one training set CV (per analyte) from the same file. 17

20 This assures random sampling of your training set standards and helps remove bias and deterministic noise. 4. Use CVs that span a large range of concentrations. Any calibration curve is only valid for the range over which one calibrates; most calibration curves fail to predict standards well when they have to extrapolate. Try varying stimulation parameters to generate cyclic voltammograms of varying intensities. Try your best to generally evenly space them out over your concentration range so one CV doesn t dominate the calibration curve s predictive ability. 5. Try your best to collect training set standards in the same location as you measure from. The shapes of ph change CVs are highly dependent on an electrode s local environment (see Takmakov et al, Anal. Chem., 2010, 82, ). 6. Carbon is a very dynamic electrode material and every electrode is slightly different. This causes slight variations in redox potentials and peak shapes. Therefore, do not use training set standards from one electrode to predict concentration information from data measured on another electrode. Sanity Checking When we perform a calibration and predict unknown concentrations, how do we know we are right? It is very difficult to ascertain the validity of absolute concentration values we obtain. To help us out, we can include several checks to make sure that our calibration curve is capable of measuring concentration values appropriately, our calibration curve makes chemical sense, and our calibration curve is applicable to predict concentration values in our unknown measurement. After pressing the Train button, you will see traces appear in two check plots: Score Plot and K Matrix CV. The Score Plot shows two colored lines, with a scatter plot of colored squares around each. This is a condensed version of PCR s calibration curve. Each square represents one training set standard. The more nearly the colored squares lie on the line, the more nearly the standards for each analyte are linearly related to each other as their concentrations increase. There are two important aspects to keep in mind when constructing a calibration curve. First, the calibration curve must be able to predict concentration values appropriately. Second, all calibration standards should equally contribute to a calibration curve s predictive capability. Stated another way, no one standard should have undue influence to leverage an entire calibration curve. By its very nature, a calibration curve is like a see-saw where points on either end tend to carry more weight in influencing its shape. 18

21 Therefore, while it is important that all points on a calibration curve be accurate, we have different levels of error tolerance depending on where they are located. Here, we include a check to rule out any poor outlier standards. If any point is too far off a line in the Score Plot, a little warning will appear below the graph, entitled "Questionable items." This will alert you to the identity of a training set CV that lies far from the pack. We recommend that you delete that item from the list and find a better sample to take its place. K Matrix CV tells you, "This is what I think a <choose your analyte> CV looks like, based on the training data you have given me." It is a mathematical solution to the question What must a pure CV look like to give predict these given concentration values from these training set CVs? Take a look. If you are analyzing for dopamine, does that CV actually look like dopamine? If not, go back and re-examine your training set. To proceed forward without doing so is likely to produce nonsense. For a full discussion of the score plot, outlier detection, and calculation of the K matrix, see Keithley and Wightman, ACS Chem. Neurosci., 2011, 2, The Concentrations Screen Press the Concentrations button to calculate concentrations and move to the Concentrations screen. Here you see a layout of seven graphs. Starting at the upper left and going from left to right they are: The color plot, just as it was on the main screen. Concentration plots, calculated for two analytes. The analyte label at the bottom of each graph is a menu, so if you have more than two analytes you may use this menu to choose which one is displayed in each graph. Residual, a measure of energy left in the data after the CV patterns of the recognized analytes have been subtracted out. The graph displays changes in the sum of the squared currents of the unknown data not represented by the relevant principal components of the training set with time. If the white trace ever crosses the red line, the concentration calculations for this view are statistically unreliable. You may want to exclude this snippet, or to find and remove a glitch. Alternatively, it means your training set is not representative of your unknown data being predicted. If several snippets have residual values that cross the threshold, go back and augment your training set. Integral charge plots for both analytes shown in the concentration plots above. These are calculated according to Hermans et al., Anal. Chem., 2008, 80, Residual Color Plot. This is what the color plot looks like after the CV representations of the relevant principal components of the analytes have been 19

22 subtracted out. If this shows anything other than a fairly smooth plot of plain, dull noise, you may want to look at it. What is going on that you haven't analyzed? Could it be interfering with the calculation of the concentrations that you are trying to analyze? Printing and Exporting Your Data Press the Printable View button to get a copy of this screen optimized for printing: no needless gray or black backgrounds wasting ink. Once you see it and are satisfied, press the Print button to print a copy. Temporary note: at present this Print command does not bring up a print dialog hence can only print to your default printer. Be sure to set up your default printer appropriately beforehand. Press Export all concentrations to get an export file of the calculated concentration traces for each snippet. When viewing basic data, press Export view to get an export file of the data currently seen on the screen. When viewing averages (see next section), press Export averages to get an export file of the average traces and the average color plot. Press Export progressions to get an export of the three-dimensional progression graphs (a separate file for each analyte). 20

23 5. Data Averaging When used appropriately, signal averaging is a great tool to enhance your signal-tonoise ratio. With HDCV you can produce average color plots and concentration profiles after processing your data through Quick Concentrations or PCR. Of course if you wish to average without doing a concentration conversion, you can use the Quick Concentration feature with a calibration factor of 1. Three types of data are supported for averaging: multiple files, multiple runs and snippets. Each is handled slightly differently by HDCV. Averaging Files Pressing Averages will allow you to choose which files to include in the average. These files should of the same length and collected with the same acquisition parameters. They do not have to be sequential, and you do not have to include the file open on the main Analysis screen in the average. This file will act as a template for averaging, however. Use it to set the filtering and cursor parameters that will be applied to every file included in the average. The only exception to this is deglitching. If you have saved your deglitching in.aly files, Analysis will pull this information individually for each file averaged. Averaging Runs Buttons in the middle of both concentration screens will allow you to toggle through each run of the loaded file. As you view each run, you may choose to accept or reject them individually for averaging. You may also flag runs of interest, and later choose to only include or exclude flagged files in the average. Indicators on the screen will track how many runs have been accepted, rejected, and flagged. For multiple run files, averaging takes filtering settings and cursor placements from the last run viewed in the main screen. Deglitching information is again individually remembered for each run. Averaging Snippets Much like multiple run files, the toggle buttons below the concentration versus time trace will allow you navigate through generated snippets. After designating which snippets you wish to include, exclude or flag, pressing Averages will prompt an intermediary dialog box. Here you can realign the snippets around a different digital 21

24 time stamp and/or adjust the timing around the event of interest. Be careful! You should probably choose a time slice somewhat smaller than your snippet. Supposing your snippets are built with 5 seconds before and after the defining event. In a snippet where the rat delayed a while before pressing the lever, if you ask for averages with 5 seconds before and after Lever Press, 5 seconds after Lever Press will run beyond the end of the snippet. The averaging machinery cannot handle such cases. Once averaging is complete, you will see three statistics beside the Averages button: Used, No fit, and No event. No Event represents snippets in which the chosen signal (e.g. Lever Press) did not occur at all. No Fit represents the case discussed above -- the time included in the average would run beyond the end of the snippet. Used represents the good snippets that were included. If you get any significant number in No Fit, you should go back and trim back the requested time for your average. in the average. The Averages Screen The averaging process may take a few seconds. When complete, the screen will change, and you will see a prominent message "Showing Averages." The graphs displayed now are as follows: The color plot, displaying an average of the color plots of all the accepted data sets. A concentration plot for each analyte, displaying concentration vs. time, averaged over all accepted data sets. A progressions graph for each analyte. These three-dimensional plots display concentration vs. time and versus file, run, or snippet. Thus each concentration graph that you would have seen while going through each data set in the previous step, is a cross-section of this three-dimensional plot. A cross-section in the other direction shows a particular moment in time, relative to the start of the file, run or snippet, traced through all data sets included. To see which data sets have been included in the averages, select Show. When you have seen all that you need to see of these averages, you can press Return to Data to return to the previous view. 22