My proteomics data must be run in factorial (factor 1: 4 bacteria inoculated on the plant and factor 2: plant with and without damage from larval herbivory). I expected to run in Two-way ANOVA but it generates very small numbers of significant proteins and I believe it is due to this limitation “Due to computational limitations, only the top 200 features (based on variance) will be tested. For analysis of all features or for unbalanced designs, we suggest using our Linear Models with Covariance Adjustment approach.”
I need to run all factors 1 and 2 together. In this case for the metaboanalyst PRO version will this limitation of 200 top features also exist for Two-way ANOVA or is this a limitation of the analysis?
For MetaboAnalyst Pro, we support top 500 features for now. We are doing more testing and may cover all features in the coming future if necessary based on users’ feedback.
We strongly recommend using limma (linear modeling) approach rather than conventional two-way ANOVA. Not only because it is computationally efficient for big data, it is more because it is statistically more robust when there are many more features with very small number of samples/replicates (3 ~ 5 per group). By using empirical Bayesian approach, limma has been to be able to improve the performance significantly while most (>90%!) of significant features identified by traditional ANOVA are false positives.