Bioinformatics

Should p-values after model selection be multiple testing corrected?

I was recently comparing different likely models (each was a different time profile) for each gene in time-series RNA-seq data. Since I did not have simple nested models, I was forced to use (as the simplest option) the Akaike Information Criterion (AIC) (I could have used the Bayesian Information Criterion as well) to select the “best” model. In the analysis of genomic data, the next step is typically thresholding the corrected p-values (i.

Combining p-values from multiple tests on the same data - I

It is common to encounter situations, where one has data from the same assay from different labs or sources, or data from different assays all targeting a phenomenon. One then proceeds to test a desired hypothesis on the basis of these multiple datasets. The only challenge in doing this is the manner in which the different datasets can be combined in a statistically appropriate way. Since in the biological context, it is rather difficult to quantify the quality of different datasets.