OmicsFusion is a web application which can perform univariate regression and a number of modern regression techniques and then ranks the predictors according to an importance criterion (mean rank over all the methods) so that in the visualization the top list contains the most important predictors.

Multivariate regression

The multivariate regression techniques used are: PCR (= principal component regression), PLS (partial least squares), ridge regression, lasso regression, elastic net regression, sparse PLS regression, and random forest regression (see also the reader on multivariate regression, provided). These techniques are different in the way they deal with the dimensionality, in the way they deal with correlated variables, and in whether or not they carry out variable selection (so: reduce the number of predictors to a smaller set of predictors).

Cross validation

Usually in these procedures, one or more parameters of the procedure need to be chosen by the user or by an automated optimization procedure. In OmicsFusion this is done by a tenfold cross-validation procedure: the data set is split into two parts: 90% of the samples are the training data, 10% are the test data. Models with different values are run using the 90% and the result is tested on the remaining 10%. This is done for revolving sets of 90% vs. 10% in a way that each sample is in the 10% exactly once per ten cross-validation rounds.