# Perform a Generalised Semi Linear Canonical Correlation Analysis

### Description

Performs (Generalised) Semi Linear Canonical Correlation Analysis, i.e. computes the canonical correlation between a data matrix and a nonlinear function of time. SLCCA is extended by allowing parameters to vary by a treatment factor and allowing adjustment for covariates. Shortcuts are provided for PK/PD models suitable for analysing EEG data.

### Usage

1 2 3 4 5 |

### Arguments

`Y` |
a data matrix. |

`formula` |
either a nonlinear model formula specifying a function
of |

`time` |
a vector of time values corresponding to the rows of |

`subject` |
an optional factor grouping the rows of Y. If specified, a separate set of coefficients of Y will be estimated for each subject. |

`global` |
if |

`treatment` |
an optional factor (nested within |

`ref` |
the reference level of |

`separate` |
if |

`partial` |
a linear model formula specifying covariates to
partial out of the CCA analysis (may be |

`data` |
an optional data frame in which to evaluate the variables
in |

`subset` |
an optional logical or numeric vector specifying a subset of observations to be used in the fitting process. |

`global.smooth` |
controls the smoothing of |

`subject.smooth` |
controls the smoothing of |

`pct.explained` |
a scalar between 0 and 1 indicating the desired
minimum percentage of variance explained by the SVD approximation
when |

`start` |
a named list of starting values for the parameters. Each
element must have the same length as the number of non-reference
levels of |

`method` |
the method to be used by |

`lower, upper` |
bounds on the nonlinear parameters. |

`...` |
arguments passed on to |

### Details

The function fits the following model:

*
Y(t) * A = X(t, theta) * B*

where *Y(t)* is a data
matrix with rows of observations recorded at times
*t*, *A* is a vector of loadings,
*X(t, theta)* is a matrix with columns
containing a nonlinear function with unknown parameters for each
non-reference treatment level, and *B* is a
vector of coefficients.

The parameters *A, theta* and
*B* are estimated to optimise the correlation
between the left- and right-hand sides of the model.

If `partial`

specifies a matrix of covariates, *G*, to be
partialled out, then the canonical correlation analysis is based on
the residuals from the multivariate linear models
`lm(Y ~ 0 + G)`

and
`lm(X ~ 0 + G)`

When `partial = ~1`

this is equivalent to centering the columns
of *Y* and *X*.

The nonlinear function defining *X* may be specified by a
character string for two pharmacokinetic/pharmacodynamic models:

`"Double Exponential"`

`~ exp(-time/exp(K1))-exp(-time/exp(K2))`

`"Critical Exponential"`

`~ time*exp(-time/exp(K1))`

The data matrix `Y`

may be pre-smoothed to remove global
artefacts. This is achieved by approximating the data from the
number of SVD roots specified by `global.smooth`

. If the
number of roots is not specified explicitly, *k* roots are
selected such that
*lambda_k/lambda_1 >0.001* and
*
(sum_{j=1}^k lambda_j)/(sum_{j=1}^r lambda_j) >= p*
where *k = 1, ..., r*, *r* is the rank of
`Y`

, *lambda_1, lambda_2,
...* are the eigenvalues of the variance matrix and *p* is
given by `pct.explained`

.

The data are then analysed for each level of `subject`

separately
and the data may again be smoothed at this level to remove local
artefacts as specified by `local.smooth`

.

The coefficients of the left-hand side matrix are sometimes referred to as loadings and are described in Brain et al (2011) as signatures.

The projected values *Y(t) * A*
and *X(t, theta) * B* are
known as the y scores and x scores respectively. In this context, they
can be thought of as the (projected) observed and fitted values.

Signatures, observed and fitted values can be visualised using
`plot.gslcca`

.

### Value

An object of class `"gslcca"`

, which is a list with components

`call` |
the call to |

`ycoef` |
a matrix of the coefficients of the left-hand side matrix, with one column per subject. |

`xcoef` |
a matrix of the coefficients of the left-hand
side matrix, with one column per subject if |

`yscores` |
a vector of y scores for all subjects. |

`xscores` |
a vector of x scores for all subjects. |

`subject` |
the variables specified by |

`treatment` |
the variables specified by |

`time` |
the variables specified by |

`ref` |
the reference level of |

`nonlinear.parameters` |
a matrix of the estimated nonlinear parameters, with one column per subject. |

`y` |
a list giving the (smoothed) left-hand side matrix for each subject,
after partialling out any covariates specified by |

`x` |
a list giving the right-hand side matrix for each subject,
after partialling out any covariates specified by |

`global.smooth` |
the number of roots used in global smoothing. |

`subject.smooth` |
the number of roots used for subject-level smoothing. |

`pct.explained` |
the percentage of variance explained by the subject-level approximation. |

`opt` |
a list of output from |

### Author(s)

Foteini Strimenopoulou and Heather Turner

### References

Brain, P., Strimenopoulou, F. and Ivarsson, M. (2011). Analysing electroencephalogram (EEG) data using Extended Semi-Linear Canonical Correlation Analysis. Submitted.

### See Also

`readSpectra`

and `bandSpectra`

for reading in spectra
and aggregating over frequency bands.

`plot.gslcca`

, `varySmooth`

and `gslcca-misc`

for functions to plot `gslcca`

results, check sensitivity with regards to
smoothing and access components of `"gslcca"`

objects.

### Examples

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 | ```
data(clonidine)
### Fit separate Double Exponential models for each rate,
### with amplitude varying by treatment
result <- gslcca(spectra, "Double Exponential",
time = Time, subject = Rat, treatment = Treatment,
subject.smooth = TRUE, pct.explained = 0.96,
data = clonidine)
## drug signature
plot(result, "signature")
## projected values
plot(result, "fitted")
## projected values
plot(result, "scores")
``` |

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker. Vote for new features on Trello.