Kernel-based classification and regression methods have been used successfully in a wide variety of
biological applications. The previously outlined Kernel-based Orthogonal Projections to Latent Structures
(K-OPLS) method features enhanced interpretational capabilities compared to alternative kernel-based
methods (e.g. Support Vector Machines) by allowing detection of unanticipated systematic variation
in the data such as instrumental drift, batch variability or otherwise unexpected biological variation.
PACKAGE FEATURE SUMMARY
The K-OPLS package provides the following functionality for MATLAB and R:
Estimation (training) of K-OPLS model.
Prediction of new data using the estimated K-OPLS model in step (1).
Cross-validation functionality to estimate the generalization error of a K-OPLS model.
Optimization of kernel parameters using grid search or simulated annealing.
Model statistics, including the explained variation of X (R2X) and Y (R2Y) as well as prediction
statistics over cross-validation for regression (Q2Y, which is inversely proportional to the
generalization error) and classification (sensitivity and specificity measures).
Plot functions to visualize e.g. score vectors as well as model statistics.
