Improving Reproducibility in Microarray Studies
Author Information
Author(s): Shi Leming, Jones Wendell D, Jensen Roderick V, Harris Stephen C, Perkins Roger G, Goodsaid Federico M, Guo Lei, Croner Lisa J, Boysen Cecilie, Fang Hong, Qian Feng, Amur Shashi, Bao Wenjun, Barbacioru Catalin C, Bertholet Vincent, Cao Xiaoxi Megan, Chu Tzu-Ming, Collins Patrick J, Fan Xiao-hui, Frueh Felix W, Fuscoe James C, Guo Xu, Han Jing, Herman Damir, Hong Huixiao, Kawasaki Ernest S, Li Quan-Zhen, Luo Yuling, Ma Yunqing, Mei Nan, Peterson Ron L, Puri Raj K, Shippy Richard, Su Zhenqiang, Sun Yongming Andrew, Sun Hongmei, Thorn Brett, Turpaz Yaron, Wang Charles, Wang Sue Jane, Warrington Janet A, Willey James C, Wu Jie, Xie Qian, Zhang Liang, Zhang Lu, Zhong Sheng, Wolfinger Russell D, Tong Weida
Primary Institution: National Center for Toxicological Research, US Food and Drug Administration
Hypothesis
The reproducibility of differentially expressed gene lists in microarray studies can be improved by using fold change ranking combined with a non-stringent p-value cutoff.
Conclusion
Using fold change ranking with a non-stringent p-value cutoff leads to more reproducible lists of differentially expressed genes in microarray studies.
Supporting Evidence
- The study found that using fold change as a ranking criterion improved the reproducibility of DEG lists.
- Inter-site comparisons showed that fold change ranking consistently yielded higher percentages of overlapping genes.
- The results indicated that the more stringent the p-value threshold, the less reproducible the DEG list became.
Takeaway
This study shows that to get reliable lists of important genes from experiments, scientists should look at how much the genes change (fold change) and not just how statistically significant they are.
Methodology
The study analyzed data from the MicroArray Quality Control (MAQC) project, comparing different gene selection methods and their impact on reproducibility.
Potential Biases
Potential biases may arise from the reliance on specific statistical methods that could affect the generalizability of the findings.
Limitations
The study primarily focused on specific statistical methods and may not encompass all possible approaches to gene selection.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website