Predict survival summaries from legacy Cox-PLS fits

These methods extend stats::predict() for Cox models fitted with the original PLS engines exposed by coxgpls(), coxsgpls(), and their deviance-residual or kernel variants. They provide access to latent component scores alongside linear predictors and risk estimates, ensuring consistent behaviour with the newer big-memory solvers.

Usage

# S3 method for class 'coxgpls'
predict(
  object,
  newdata = NULL,
  type = c("link", "risk", "response", "components"),
  comps = NULL,
  coef = NULL,
  ...
)

# S3 method for class 'coxgplsDR'
predict(
  object,
  newdata = NULL,
  type = c("link", "risk", "response", "components"),
  comps = NULL,
  coef = NULL,
  ...
)

# S3 method for class 'coxsgpls'
predict(
  object,
  newdata = NULL,
  type = c("link", "risk", "response", "components"),
  comps = NULL,
  coef = NULL,
  ...
)

# S3 method for class 'coxsgplsDR'
predict(
  object,
  newdata = NULL,
  type = c("link", "risk", "response", "components"),
  comps = NULL,
  coef = NULL,
  ...
)

# S3 method for class 'coxspls_sgpls'
predict(
  object,
  newdata = NULL,
  type = c("link", "risk", "response", "components"),
  comps = NULL,
  coef = NULL,
  ...
)

# S3 method for class 'coxDKgplsDR'
predict(
  object,
  newdata = NULL,
  type = c("link", "risk", "response", "components"),
  comps = NULL,
  coef = NULL,
  ...
)

# S3 method for class 'coxDKsgplsDR'
predict(
  object,
  newdata = NULL,
  type = c("link", "risk", "response", "components"),
  comps = NULL,
  coef = NULL,
  ...
)

# S3 method for class 'coxDKspls_sgplsDR'
predict(
  object,
  newdata = NULL,
  type = c("link", "risk", "response", "components"),
  comps = NULL,
  coef = NULL,
  ...
)

Arguments

object: A fitted model returned by coxgpls(), coxsgpls(), coxspls_sgpls(), or any of their deviance-residual/kernel counterparts with allres = TRUE.
newdata: Optional matrix or data frame of predictors. When NULL, the training components stored in object are reused.
type: Type of prediction requested: "link" for linear predictors, "risk"/"response" for exponentiated scores, or "components" to return latent PLS scores.
comps: Optional integer vector specifying which latent components to retain. Defaults to all available components.
coef: Optional coefficient vector overriding the Cox model coefficients stored in object.
...: Unused arguments for future extensions.

Value

When type is "components", a matrix of latent scores; otherwise a numeric vector containing the requested prediction with names inherited from the supplied data.

References

Bastien, P., Bertrand, F., Meyer, N., & Maumy-Bertrand, M. (2015). Deviance residuals-based sparse PLS and sparse kernel PLS for censored data. Bioinformatics, 31(3), 397–404. doi:10.1093/bioinformatics/btu660

Bertrand, F., Bastien, P., & Maumy-Bertrand, M. (2018). Cross validating extensions of kernel, sparse or regular partial least squares regression models to censored data. https://arxiv.org/abs/1810.01005.

Examples

if (requireNamespace("survival", quietly = TRUE)) {
  data(micro.censure, package = "bigPLScox")
  data(Xmicro.censure_compl_imp, package = "bigPLScox")

  X <- as.matrix(Xmicro.censure_compl_imp[1:60, 1:10])
  time <- micro.censure$survyear[1:60]
  status <- micro.censure$DC[1:60]

  set.seed(321)
  fit <- coxgpls(
    Xplan = X,
    time = time,
    status = status,
    ncomp = 2,
    allres = TRUE
  )

  predict(fit, newdata = X[1:5, ], type = "risk")
  head(predict(fit, type = "components"))
}
#>           dim.1      dim.2
#> [1,]  0.8180843  1.0402867
#> [2,] -1.6095533  0.0997729
#> [3,]  0.1280324 -0.5407040
#> [4,] -0.1346531 -1.9098214
#> [5,]  0.8531046 -0.3185590
#> [6,]  0.2168484 -0.3475775