Raina Dong's List: Biology

• • book mark
    Calculating Permutation-Based Corrected P-values
• long-time standard test statistic for comparing two groups is the t-statistic
• • bookmark
• A drawback of the t-statistic for microarray datasets is that most experiments have only a few samples in each group
• However the t-test does fall down badly when there are outliers (values more than twice as far from the mean as all the others)
• this approach is particularly useful for microarray studies because it can be easily adapted to estimate significance levels for many genes in parallel.
• The steps in a permutation-based computation of the significance level of a test statistic are as follows:
• The idea is that if the gene is distributed similarly in both treatment and control groups, then the difference statistic (a t-statistic or any other) will appear about as big in the permuted arrangement, as in the true arrangement.
• A permutation test needs at least two groups of six samples, in order to have enough different permutations.
• balanced permutations
• The first axis indicates biological impact of the change; the second indicates the statistical evidence, or reliability of the change.
• The FDR is the expected fraction of false positives in a list of genes selected following a particular statistical procedure.
• The q-value is the smallest FDR at which a particular gene would just stay on the list of positives.
• An FDR of 5 or 10% should be acceptable for journal publication of gene lists

12 more annotations...