xcorr

The 'xcorr' analysis routine computes cross-correlograms (CCGs), auto-correlograms (ACGs), and all-way shift-predictors. It operates on spike data and continuous (float) data.

A peak at negative times means that chan1 fires before chan2. Thus, the time axis shows what chan1 does relative to a spike on chan2. (Note, the nData Viewer has the opposite convention, which is that the first channel (Channel 0) is the reference channel.)

Options:

• plot - make a CCG (or ACG) plot. Use the 'xcorr_flag', 'shift_flag', and 'diff_flag' options to chose which plots are written. 'diff' indicates the CCG minus the shift predictor.
• avg - plot average of diff, shift, or xcorr across trials. For diff, the average is taken across differences (xcorr-shift) for each group.
• avg_stat - write stats for xcorr peak.
  1. infile - name of input file
  2. n - number of plots averaged together.
  3. imax - time lag of peak
  4. max - value at peak
  5. fleft - interpolated left-half rise (first to left of peak)
  6. fright - interpolated right-half rise (first to right of peak)
  7. fwidth - width at half-height (fright - fleft)
• special_01 -
• special_02 -
• special_03 -
• ac_stat -
• noise_stat -
• rpeak_time -
• rpeak_time_excite -

Parameter definitions:

• xnorm - (frac1)
  • raw - coincidences/sec. Or, no normalization for continuous data.
  • gmsr - divide by geometric mean spike rate.
  • trig1 - spike rate of chan2 given spike on chan1 (thus, the CCG baseline should be the mean firing rate of chan2)
  • trig2 - spike rate of chan1 given spike on chan2 (thus, the CCG baseline should be the mean firing rate of chan1)
  • frac - ratio of coincidences to expected (for Poisson) coincidences, normalized by individual trial.
  • frac1 - like frac, but normalized by mean spike count.

  • zscore - for continuous data
  • zeromean - for continuous data
• shift_type - 'allway', 'periodic', ... (works with 'plot' option)
  • allway - (default) use all possible unmatched trial pairs.
  • periodic - shift spikes in trial to left and right, recompute CCG, for periodic (e.g. sinusoidal) stimuli (must set shift_period, see below).
  • simple - this option currently applies only to continuous (float) data. It computes a shift predictor by simply shifting the second channel data by one index (allowing wrap-around) and computing the cross-correlation.
• shift_period - 'pep_periodic' or [int_shift_value]
• shift_sigma - SD for Gaussian smoothing of shift-predictor.
• sigma - SD for Gaussian smoothing of xcorr.
• lag - Maximum lag time for plot (sampling units).
• centerflag - replace central value in CCG with average of two neighboring values.
• scenterflag - replace central value in shift-predictory with average of two neighboring values.
• weight_flag - if 1, use weighted average where weight is n times pow(GMSR,weight_power).
• weight_power - The exponent used for computing weights, pow(GMSR,weight_power), by default is 0.0.
• mintrials - minimum number of trials for valid condition.
• mintotspikes - miniumum total number spikes (ignored by some analyses.
• chan1 - first channel name.
• chan2 - second channel name, set equal to chan1 for ACG.
• xcorr_flag - plot the CCG without subtracting the shift predictor.
• shift_flag - plot the shift predictor.
• diff_flag - plot the CCG after subtracting the shift predictor.
• tau0 - Used with 'ac_stat', starting lag for SD calculation.
• taun - Number of consecutive lags for SD calculation (for 'ac_stat')
• exciteflag -
• suppress_win0 - Suppress spikes on chan2 relative to those on chan1 beginning at this time. This will induce negative correlation overall.
• suppress_winn - Suppress spikes on chan2 for this epoch following suppress_win0.

  Efficacy and Contribution

• stat_flag_ec - (0) 1-compute efficacy and contribution statistics. The "xnorm" must be "raw". You must set a window for integrating the CCG peak using "win0" and "winn". The following will be written to a file named after the 'outfile' but with an additional '.stat' extension:
  1. cwd - current working directory
  2. infile - input file name
  3. outfile - output file name
  4. sr1 - firing rate of first channel (spikes/sec)
  5. sr2 - firing rate of second channel (spikes/sec)
  6. xarea - integral of raw (CCG - Shift) curve within window (excess coincidence rate)
  7. strength = 100 * xarea/[(sr1+sr2)/2]     (Dan et al 1998)
  8. toyama = 100 * (xarea/sr1 + xarea/sr2)/2     (Toyama et al 1981)
  9. simple efficacy = 100 * xarea/sr1
  10. simple contribution = 100 * xarea/sr2
  11. x = xarea/sampling + sp1*sp2 (see next)
  12. sp1 = sr1/sampling
  13. sp2 = sr2/sampling
  14. efficacy = (x/sp1 - sp2) / (1 - sp1 - sp2 + x)
  15. contribution = efficacy * sp1 / sp2
• win0 - (0) start time lag of integration window (sampling units).
• winn - (10) duration of integration window (sampling units).

Example:

#
#  Compute cross-correlation and shift-predictor, avarage over all trials.
#
xcorr
all
plot

chan1 unit0        # 'unit0' is name of first channel of data
chan2 unit1        # 'unit1' is name of first channel of data

sampling 1000.0    # time units are milliseconds

start 200          # start of analysis window
period 1800        # length of analysis window
lag 200            # plot CCG from plus/minus this time lag

sigma 2.0          # SD of Gaussian for smoothing the CCG (not shift predictor)

xcorr_flag 1       # Write the raw CCG to the output file
shift_flag 1       # Write the shift predictor to the output file
diff_flag 1        # Write the difference, CCC - shift, to output file

xnorm gmsr         # Use the 'geometric mean spike rate' normalization

Example:

#
#  Cross-correlation and shift-predictor for continous data
#
xcorr
group 1 direction
plot

chan1 v1_opp1       # name of first channel of data
chan2 mt            # name of second channel of data

sampling 1000.0     # time units are milliseconds

start 50            # start of analysis window
period 950          # length of analysis window
lag 200             # plot CCG from plus/minus this time lag

sigma        2.0    # SD of Gaussian for smoothing the CCG (not shift predictor)
shift_sigma  0.0    # SD of Gaussian for smoothing the shift predictor

shift_type  simple  # 'simple' is the only choice for float data.

xcorr_flag 1        # Write the raw CCG to the output file
shift_flag 1        # Write the shift predictor to the output file
diff_flag 1         # Write the difference, CCC - shift, to output file

xnorm  zscore       # 'zscore', 'zeromean' or 'raw'

Example:

#
#  Cross-correlation for coherence series from Newsome Data.
#
xcorr
group 1 coherence
special_01

start 300
period 1700
lag 200
xnorm gmsr
centerflag 1
mintrials 4
weight_flag 1
weight_power 0.0

condition
point_count_min unit0 300 1700 4
point_count_min unit1 300 1700 4