Independent Filters - Anderson-Lab/ComplementaryDomainPrioritization GitHub Wiki

##Required File The independent filtering uses a transposed data frame (rows are samples, columns are variables), and is meant to be used following prioritization. Depending on which prioritization strategy example you have followed, you will have one of these files

KEGGprioritized.data
Wikiprioritized.data
TFprioritized.data

Independent Filtering

Independent filtering is a way to increase the power of analysis in an unsupervised manner by using either variance or mean abundance of features in a data set to rank and then remove features. This is discussed extensively by Bourgon et al. and in the accompanying paper to this package.

For either variance or abundance based filtering, a θ can be specified, which is the target number of ranked variables; otherwise, it is the top θ fraction of ranked variables. A typical approach would be use θ = 0.5, which would provide a stringent filter on the data. If comparisons are required and the desire is to keep a specific number of genes for analysis, θ = 25000 (or some number) may be used. For this example we will use θ = 0.5 in order to filter out 50% of the KEGG prioritized data.

KEGGprioritized.50varfiltered.data = overall.var.filter(KEGGprioritized.data,0.5)
KEGGprioritized.50meanfiltered.data = overall.mean.filter(KEGGprioritized.data,0.5)

This results in either variance filtered or mean abundance filtered data frames. Again, there is a $label which can be used to assign class labels. As discussed in Bourgon et al., care should be taken about choice of filter based on downstream analysis (for example, using a variance filter along with a moderated t-test likely decreases power by affecting variance distributions).