Orthogonal-Partial Least Squares (O-PLS)

This function fits Orthogonal-Partial Least Squares (O-PLS) models for regression (R) or discriminant analysis (DA). The optimal number of components is determined automatically using automated stop criteria based on statistical cross validation indices.

opls(
  X,
  Y,
  center = TRUE,
  scale = "UV",
  cv = list(method = "k-fold_stratified", k = 7, split = 2/3),
  maxPCo = 5,
  plotting = TRUE
)

Arguments

X	Numeric input matrix or dataframe (usually measurements obtained through NMR spectroscopy or mass spectrometry) with each row representing an observation and each column a metabolic feature.
Y	Response vector or matrix with same length or number of columns than rows in X, respectively. Y have multiple columns.
center	Logical value (TRUE or FALSE) indicating if features should be mean centered.
scale	Desired scaling method (currently only no or unit variance scaling (UV) implemented).
cv	Named list of cross-validation paramters to derive the optimal number of components: method, one of 'k-fold', 'k-fold_stratified', 'MC', 'MC_balanced' (see Details), split: fraction of observations used for model training, k: k-fold paramter, ie., the number of cross-validation sets. The latter depends on the number of observations in X but typically takes a value between 3 and 9.
maxPCo	The maximum number of orthogonal components (in case stop criteria fail).
plotting	Logical value (TRUE or FALSE) indicating if model parameters (R2X, Q2, etc) should be visualised once the model is trained.
t_pred	Parameter specifying the maximum number of predictive components (needed only for multi-factor Y)

Value

This function returns an metabom8 object (S4).

Details

Models are fully statistically validated, currently only k-fold cross validation (CV) and class-balanced k-fold cross validation is implemented. Further extensions, e.g. Monte-Carlo CV, are work in progress. Although the algorithm accepts three and more levels as Y, model interpretation is more straightforward for pairwise group comparisons.

References

Trygg J. and Wold, S. (2002) Orthogonal projections to latent structures (O-PLS). Journal of Chemometrics, 16.3, 119-128.

Geladi, P and Kowalski, B.R. (1986), Partial least squares and regression: a tutorial. Analytica Chimica Acta, 185, 1-17.

Author

torben.kimhofer@murdoch.edu.au

Examples

data(covid)
model=opls(X, Y=an$type)
#> Performing discriminant analysis.
#> Error in .checkDimXY(X, y_check[[1]]): Error: Dimensions of input X and Y do not match.
plotscores(model, an=list(Class=an$type, Clinic=an$hospital, id=1:nrow(an)), pc=c(1,2))
#> Error in grepl("OPLS", class(obj)[1]): object 'model' not found