Unsupervised approaches: potential outliers can be identified from PCA or heatmaps through visualization. The scores plot can be used to identify sample outliers, while the loadings plot can be used to identify feature outliers. The potential outlier will distinguish itself as the one located far away from the major clusters formed by the remaining.
Supervised approaches: Random Forest method (Statistical Analysis module) also provide information and visualization on potential outliers; the Prob. View (multivariate biomarker analysis) is also very useful for outlier detection (samples already predicted in the wrong group)
To deal with outliers, the first is to check if those samples / features are measured properly. In many cases, outliers are the result of operational errors during analytical process. If those values cannot be corrected, they should be removed from analysis. MetaboAnalyst provides DataEditor to enable easy removal of sample/feature outliers. Please note, you may need to re-normalize the data after outlier removal.