Robust universal scoring function for threshold free gene set analysis: Supplementary text

Short note: This is a supplementary text for the manuscript with the same title. The additions are the parts that handle stability under randomizations, pairwise p-value comparison of GSZ with other tested scoring functions, the most of the materials and methods, and derivation of E(S|N) and D(S|N) required for the derivation of the GSZ-score. This text represent five major findings: (a) Stability of the regularized GSZ-score over different threshold positions and instability of the KS and the modKS over the same threshold positions, (b) stability of the GSZ-score over different class sizes in row randomizations, (c) instabilities observed in the analysis methods when looking at the column randomizations, (d) stability of the results from the compared methods after using our normalization with randomized datasets, (e) better performance shown by GSZ-score in pairwise comparisons especially with biologically relevant classes.