gsspsth: Smooth Peri Stimulus Time Histogram Related Functions and...
In STAR: Spike Train Analysis with R

Description Usage Arguments Details Value Note Author(s) References See Also Examples

Function gsspsth and gsspsth0 compute a smooth psth, while method print.gsspsth and print.gsspsth0 print and summary.gsspsth or summary.gsspsth0 summarize the gssanova / gssanova0 objects contained in the returned gsspsth or gsspsth0 objects, plot.gsspsth or plot.gsspsth0 plot them and simulate.gsspsth or simulate.gsspsth0 simulate data from fitted objects.

gsspsth(repeatedTrain, binSize = 0.025, plot = FALSE, ...)
gsspsth0(repeatedTrain, binSize = 0.025, plot = FALSE, ...)
## S3 method for class 'gsspsth'
print(x, ...)
## S3 method for class 'gsspsth0'
print(x, ...)
## S3 method for class 'gsspsth'
summary(object, ...)
## S3 method for class 'gsspsth0'
summary(object, ...)
## S3 method for class 'gsspsth'
plot(x, stimTimeCourse = NULL, colStim = "grey80",
           colCI = NULL, xlab, ylab, main, xlim, ylim,
           lwd = 2, col = 1, ...)
## S3 method for class 'gsspsth0'
plot(x, stimTimeCourse = NULL, colStim = "grey80",
           colCI = NULL, xlab, ylab, main, xlim, ylim,
           lwd = 2, col = 1, ...)
## S3 method for class 'gsspsth'
simulate(object, nsim = 1, seed = NULL, ...)
## S3 method for class 'gsspsth0'
simulate(object, nsim = 1, seed = NULL, ...)

`repeatedTrain`	a `repeatedTrain` object or a list which can be coerced to such an object.
`binSize`	the bin size (in s) used to generate the observations on which the gss fit will be performed. See details below.
`plot`	corresponding argument of `hist`. Should a plot be generated or not?
`object`	a `gsspsth` or a `gsspsth0` object.
`x`	a `gsspsth` or a `gsspsth0` object.
`stimTimeCourse`	`NULL` (default) or a two elements vector specifying the time boundaries (in s) of a stimulus presentation.
`colStim`	the background color used for the stimulus.
`colCI`	if not `NULL` (default) a confidence band is plotted with the specified color; two dashed lines are plotted otherwise.
`xlim`	a numeric (default value supplied). See `plot`.
`ylim`	a numeric (default value supplied). See `plot`.
`xlab`	a character (default value supplied). See `plot`.
`ylab`	a character (default value supplied). See `plot`.
`main`	a character (default value supplied). See `plot`.
`lwd`	line width used to plot the estimated density. See `plot`.
`col`	color used to plot the estimated density. See `plot`.
`nsim`	number of `repeatedTrain` objects to simulate. Defaults to 1.
`seed`	see `simulate`.
`...`	in `gsspsth`, respectively `gsspsth0`, the ... are passed to the internally called `gssanova`, repectively `gssanova0`. In `plot.gsspsth` and `plot.gsspsth0` they are passed to `plot` which is called internally. They are not used otherwise.

gsspsth calls internally gssanova while gsspsth0 calls gssanova0. See the respective documentations and references therein for an explanation of the differences. For both gsspsth and gsspsth0, the raw data contained in repeatedTrain are pre-processed with hist using a bin size given by argument binSize. This binSize should be small "enough". That is, the rate of the aggregated train created by collapsing the spike times of the different trials onto a single "pseudo" spike train, should not change too much on the scale of binSize (see Ventura et al (2002) Sec. 4.2 p8 for more details). Argument nbasis of gssanova called internally by gsspsth is set to the number of bins of the histogram resulting from the preprocessing stage.

simulate.gsspsth and simulate.gsspsth0 perform exact simuations of inhomogenous Poisson processes whose (time dependent) rate/intensity function is accessible through the componenent lambdaFct of objects of class gsspsth and gsspsth0. The inhomogenous Poisson processes are simulated with the thinning method (Devroye, 1986, pp 253-256).

When plot is set to FALSE in gsspsth, repectively gsspsth0, a list of class gsspsth, respectively gsspsth0, is returned and no plot is generated. These list have the following components:

`freq`	a vector containing the instantaneous firing rate in the middle of the "thin" bins used for preprocessing.
`ciUp`	a vector with the upper limit of a pointwise 95% confidence interval. Check `predict.ssanova` for details.
`ciLow`	a vector with the lower limit of a pointwise 95% confidence interval.
`breaks`	a vector with 2 elements the ealiest and the latest spike in `repeatedTrain`.
`mids`	a numeric vector with the mid points of the bins.
`counts`	a vector with the actual number of spikes in each bin.
`nbTrials`	the number of trials in `repeatedTrain`.
`lambdaFct`	a function of a single time argument returning the estimated intensity (or instantaneous rate) at its argument.
`LambdaFct`	a function of a single time argument returning the integrale of estimated intensity (or instantaneous rate) at its argument. That is, the integrated intensity. `integrate` is used by this function.
`call`	the matched call.

When plot is set to TRUE nothing is returned and a plot is generated as a side effect. Of course the same occurs upon calling plot.gsspsth with a gsspsth object argument or plot.gsspsth0 with a gsspsth0.

print.gsspsth returns the result of print applied to the gssanova object generated by gsspsth and stored in the environment of both lambdaFct and LambdaFct. The same goes for print.gsspsth0.

summary.gsspsth returns the result of summary.gssanova applied to the gssanova object generated by gsspsth and stored in the environment of both lambdaFct and LambdaFct. The same goes for summary.gsspsth0.

simulate.gsspsth and simulate.gsspsth0 return a repeatedTrain object if argument nsim is set to one and a list of such objects if it is greater than one.

Most of the components of the list returned by gsspsth and gsspsth0 are not of direct interest for the user but they are used by, for instance, reportHTML.repeatedTrain.

Christophe Pouzat christophe.pouzat@gmail.com

Gu C. (2002) Smoothing Spline ANOVA Models. Springer.

Ventura, V., Carta, R., Kass, R. E., Gettner, S. N. and Olson, C. R. (2002) Statistical analysis of temporal evolution in single-neuron firing rates. Biostatistics 3: 1–20.

Kass, R. E., Ventura, V. and Cai, C. (2003) Statistical smoothing of neuronal data. Network: Computation in Neural Systems 14: 5–15.

Devroye Luc (1986) Non-Uniform Random Variate Generation. Springer. Book available in pdf format at: http://cg.scs.carleton.ca/~luc/rnbookindex.html.

psth, plot.psth, gssanova, gssanova0, summary.gssanova, summary.gssanova0, reportHTML.repeatedTrain, simulate

## Not run: 
## Get the e070528citronellal data set into workspace
data(e070528citronellal)
## Compute gsspsth without a plot for neuron 1
## using a smmothing spline with gssanova0, and default bin size of 25 ms.
n1CitrGSSPSTH0 <- gsspsth0(e070528citronellal[[1]])
## plot the result
plot(n1CitrGSSPSTH0,stim=c(6.14,6.64),colCI=2)
## get a summary of the gss fit
summary(n1CitrGSSPSTH0)
## Now take a look at the observation on which the gss
## was actually performed
plot(n1CitrGSSPSTH0$mids,n1CitrGSSPSTH0$counts,type="l")
## Add the estimated smooth psth after proper scaling
theBS <- diff(n1CitrGSSPSTH0[["mids"]])[1]
Y <- n1CitrGSSPSTH0$lambdaFct(n1CitrGSSPSTH0$mids)*theBS*n1CitrGSSPSTH0$nbTrials
lines(n1CitrGSSPSTH0$mids,Y,col=4,lwd=2)

## check the (absence of) effect of the pre-binning by using a smaller
## and larger one, say 10 and 75 ms
n1CitrGSSPSTH0_10 <- gsspsth0(e070528citronellal[[1]],binSize=0.01)
n1CitrGSSPSTH0_75 <- gsspsth0(e070528citronellal[[1]],binSize=0.075)
## plot the "high resolution" smoothed-psth
plot(n1CitrGSSPSTH0_10,colCI="grey50")
## add to it the estimate obtained with the "low resolution" one
Y_75 <- n1CitrGSSPSTH0_75$lambdaFct(n1CitrGSSPSTH0_10$mids)
lines(n1CitrGSSPSTH0_10$mids,Y_75,col=2,lwd=2)

## simulate data from the first fitted model
s1 <- simulate(n1CitrGSSPSTH0)
## look at the result
s1

## Do the same thing with gsspsth
n1CitrGSSPSTH <- gsspsth(e070528citronellal[[1]])
plot(n1CitrGSSPSTH,stim=c(6.14,6.64),colCI=2)
summary(n1CitrGSSPSTH)
plot(n1CitrGSSPSTH$mids,n1CitrGSSPSTH$counts,type="l")
theBS <- diff(n1CitrGSSPSTH[["mids"]])[1]
Y <- n1CitrGSSPSTH$lambdaFct(n1CitrGSSPSTH$mids)*theBS*n1CitrGSSPSTH$nbTrials
lines(n1CitrGSSPSTH$mids,Y,col=4,lwd=2)
## check the (absence of) effect of the pre-binning by using a smaller
## and larger one, say 10 and 75 ms
n1CitrGSSPSTH_10 <- gsspsth(e070528citronellal[[1]],binSize=0.01)
n1CitrGSSPSTH_75 <- gsspsth(e070528citronellal[[1]],binSize=0.075)
## plot the "high resolution" smoothed-psth
plot(n1CitrGSSPSTH_10,colCI="grey50")
## add to it the estimate obtained with the "low resolution" one
Y_75 <- n1CitrGSSPSTH_75$lambdaFct(n1CitrGSSPSTH_10$mids)
lines(n1CitrGSSPSTH_10$mids,Y_75,col=2,lwd=2)
## simulate data from the first fitted model
s1 <- simulate(n1CitrGSSPSTH)
## look at the result
s1

## End(Not run)