2 Statistical tests - cbib/DIMet GitHub Wiki

Univariate analyses

Type of analysis	Test abbreviation	Full Name	Null hypothesis	Principle	Usage recommendation
Pairwise or Time-course	ranksum	Rank sum test (Wilcoxon’s)	Two populations have the same distribution	Based on the sum of the positions in which the values of the two conditions fall.	Unpaired samples. Equal or unequal sized samples. Any number of variables.
	Wcox	Wilcoxon’s signed rank	Differences across two matched populations are close to 0	Based on the signs of the matched differences	Paired or related -non independent- samples (check your experimental design). Any number of variables.
	MW	Mann Whitney	Two populations have the same distribution	Based on the order in which the values from the two conditions fall	Independent samples; a.k.a “the non-parametric version of the t-test”. Assumes equal variances. Any number of variables.
	KW	Kruskal-Wallis	Two populations share the same medians	Based on the difference in the group-wise order totals	Unpaired, independent samples. Works better with n 5 Any number of variables.
	BrMu	Brunner-Munzel	Two populations share the same range of values	Difference in the averaged orders of the values, normalized by the variances.	Unpaired, independent samples, being n 10. Highly similar to the Mann-Whitney test, but does not require equal variances. Any number of variables.
	disfit	Fitting of a distribution to the z-scores	Data follows the specified distribution	Identify outliers using the best fit for the ratios of geometric means of values in 2 conditions	Unpaired and paired samples. Number of variables: use it if there are thousands or hundreds of isotopologues. Do not use it for fractional contributions or total metabolite abundances if you have few (<100) metabolites.
	prm-scipy	Permutations method via scipy	Two populations are drawn from the same underlying distribution	Differences between means of subsampled values from 2 conditions	Unpaired and paired samples. Any number of variables.
Multi-group	KW	Kruskal-Wallis	Three or more populations share the same medians	Based on the difference in the group-wise order totals	Independent samples; a.k.a “the non-parametric alternative to the ANOVA test”. Works better with n 5 . Any number of variables.

Bi-variate analyses

Type of analysis	Full Name	Null hypothesis	Principle
MDV profile comparison between 2 conditions	Pearson or Spearman test	The MDV values of the two conditions are not correlated	Correlation of the MDV geometric means of the two conditions
MDV profile comparison between 2 consecutive time-points	Pearson or Spearman test	The MDV values of the two consecutive time-points are not correlated	Correlation of the MDV geometric means of two consecutive time-points
Metabolite total abundances and fractional contribution time-course profile comparison between 2 conditions	Pearson or Spearman test	The time-course values of metabolite total abundance or fractional contribution of the two conditions are not correlated	Correlation of the geometric means (computed across the time-course set of values) of the two conditions

Note: the Spearman test (spearman) is the option set by default in the DIMet's bi-variate analysis.