Integrate plotloadings result and vip result


Thanks for this simple and great package.
I’m now using PLS-DA with transcriptome and Proteome expression data.

In my data, each feature is divided into a total of five groups. For this, we performed plsda to find out which feature can explain the well-divided of these five groups and obtained a vip score using the vip function.

In my case, plsda was performed with comp=4, and the result of vip seems to give a score corresponding to each component. After that, I want to find out which group the feature with high vip score contributes to.

As a result of the function plotloadings(), the contents of the column ‘GroupContrib’ in the result datatable determine which group each feature contributes to. I wonder if it is possible to interpret the result by integrating the result of vip().
Or, I would like to get advice on how to select features with a VIP score and find out if it corresponds to five groups.

I would recommend the use of the loading values to determine the importance of your predictors rather than the VIP. Due to the formulation of VIP, it has a tendency to give infrequently selected (sub-optimal) features a high score.

As you’ve done, looking at the output of plotLoadings is the way to go. I’m not sure there would be any use in integrating the VIP into this information. VIP measures the contribution of each feature to the overall variance of the data. With PLS-DA, covariance is the measure used. Hence, its integration may lead to incorrect interpretations.

1 Like