Lecture 8 Differential Expression

Propability value

Propability of obtaining test results at least as extreme assuming the null hypothesis is correct

Rejecting the Null hypothesis even though it is true

Propability to fail to reject the null hypothesis given that its false.

A test to compare if 2 means are statistically significantly different

• Null distribution is the t distribution

m1 / m2 mean values

s scale ( sum of variances of both groups normalized by sample sizes)

c scale for different group sizes

It assumes equal variance in both sample

• => Welch-test can be used instead

With small samples the t-test is only valid if Xj is normally distributed

Variance estimator is highly variable with small sample sizes

Fold-changes

Modified t-tests

Permutation procedures

• no distributional assumptinos, compute intensive, not adequate for small samples

Wilcoxon-test (distribution free, does not require normal dist)

• only take rank into account

• not adequate for small samples

By doing many many different univariate tests some null hypothesis will be rejected by chance

binary Y; transcriptomic X => t-test

binary Y; SNP data => X2-test, Fishers exact test

censored Y; transcripomic data => Wald test