GSEA p-value和FDR计算方法

最新推荐文章于 2024-04-05 15:18:40 发布

原创

最新推荐文章于 2024-04-05 15:18:40 发布 · 1.2k 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#机器学习

文章介绍了评估观察到效果显著性的两种方法：一是通过比较观察到的效果大小（ES）与随机分配表型后计算的ES_NULL分布来估计p值；二是通过多次固定表型标签的重排，调整基因集大小，计算FDR，控制假阳性率，并利用NES(S,π)的分布计算FDRq值，以确定特定显著性水平的基因集。

1，p-value计算方法

Estimating Significance. We assess the significance of an observed ES by comparing it with the set of scores ESNULL computed with randomly assigned phenotypes.

Randomly assign the original phenotype labels to samples, reorder genes, and re-compute ES(S).
Repeat step 1 for 1,000 permutations, and create a histogram of the corresponding enrichment scores ESNULL.
Estimate nominal P value for S from ESNULL by using the positive or negative portion of the distribution corresponding to the sign of the observed ES(S).

2，FDR计算方法

Multiple Hypothesis Testing.

Determine ES(S) for each gene set in the collection or database.
For each S and 1000 fixed permutations π of the phenotype labels, reorder the genes in L and determine ES(S, π).
Adjust for variation in gene set size. Normalize the ES(S, π) and the observed ES(S), separately rescal