MagiCMicroRna is presented to the user in two ways. General users are invited to use the graphical web-interface that runs locally on their computer using the R-package shiny [3]. Three source files need to be downloaded, after which MagiCMicroRna can be started using the runApp command of shiny. Advanced users that desire to tweak settings or change the analysis according to their specific needs, can also download and edit the source code for running it directly in R.
Both the web interface and source code accept two types of input files. First, the target-file links the raw data files with the experimental design and treatment groups. This file resembles to some extent the original used by López-Romero [1], but contains additional columns in order to perform the group-specific filtering (described in https://bitbucket.org/mutgx/magicmicrorna.git). Secondly, MagiCMicroRna needs the raw .txt-files obtained by Agilent Feature Extraction Software.
The MagiCMicroRna algorithm is shown in Figure 1. Briefly, it reads the target- and raw datafiles and extracts the summarized TotalGeneSignal for each miRNA. It then normalizes the data based on the algorithm that the user has chosen. Information on the normalization method to use can be found in [2]. The normalized data matrix for all samples is written to the output file Normalized_<norm.method>_TotalGeneSignal.txt. In contrast to the original package, the output file NOCtrl_exprs.txt is omitted. Instead of normalized data, as was stated by the authors [1], this file contained the raw mean signals for the first occurrence of each miRNA-probe on the array. Subsequently, the user has the option to perform group-specific or overall filtering of miRNAs that passed QC-criteria. Both approaches have their (dis-)advantages. Group-specific filtering applies the filtering criteria to a subset of samples that belong to the respective treatment and control groups. Using this approach prevents interference of aberrant samples on the miRNA-filtering of particular experimental treatments. This is especially useful when the dataset exhibits heterogeneity between groups, where particular miRNAs are only expressed in a small number of samples (e.g. a particular treatment). With overall filtering, these miRNAs would be filtered out before statistical analysis, whilst these can be very interesting to the biologist. Experimental designs that investigate independent treatment effects can greatly benefit from this more directed filtering approach. On the other hand, group-specific filtering creates output files that have different dimensions across experimental groups. In that perspective, overall filtering is recommended for experimental setups where multiple comparisons are of interest, e.g. time series, dose–response or multifactorial designs. In this way the dataset provided to subsequent statistical methods (i.e. linear modeling) does not have missing values. The user thus needs to make a careful decision about the different filtering methods.
In view of the fact that statistical analysis procedures are hard to generalize due to differences in experimental designs, we decided to leave out the linear modelling implemented in the original AgiMicroRna package and focus only on the comprehensive normalization and filtering procedures. Our normalized and filtered output files can be used directly in most statistical analysis tools.