A network-based method to evaluate quality of reproducibility of differential expression in cancer genomics studies
Metrics: PDF 1236 views | HTML 1871 views | ?
Robin Li1,2,*, Xiao Lin1,*, Haijiang Geng1, Zhihui Li1, Jiabing Li1, Tao Lu1,2, Fangrong Yan1,2
1Research Center of Biostatistics and Computational Pharmacy, China Pharmaceutical University, Nanjing, Jiangsu, China
2State Key Laboratory of Natural Medicines, China Pharmaceutical University, Nanjing, Jiangsu, China
*These authors have contributed equally to this work
Tao Lu, e-mail: firstname.lastname@example.org
Fangrong Yan, e-mail: email@example.com
Keywords: cancer genomics, gene expression, overlapping genes, pagerank, reproducibility
Received: June 23, 2015 Accepted: October 01, 2015 Published: November 09, 2015
BACKGROUND: Personalized cancer treatments depend on the determination of a patient’s genetic status according to known genetic profiles for which targeted treatments exist. Such genetic profiles must be scientifically validated before they is applied to general patient population. Reproducibility of findings that support such genetic profiles is a fundamental challenge in validation studies. The percentage of overlapping genes (POG) criterion and derivative methods produce unstable and misleading results. Furthermore, in a complex disease, comparisons between different tumor subtypes can produce high POG scores that do not capture the consistencies in the functions.
RESULTS: We focused on the quality rather than the quantity of the overlapping genes. We defined the rank value of each gene according to importance or quality by PageRank on basis of a particular topological structure. Then, we used the p-value of the rank-sum of the overlapping genes (PRSOG) to evaluate the quality of reproducibility. Though the POG scores were low in different studies of the same disease, the PRSOG was statistically significant, which suggests that sets of differentially expressed genes might be highly reproducible.
CONCLUSIONS: Evaluations of eight datasets from breast cancer, lung cancer and four other disorders indicate that quality-based PRSOG method performs better than a quantity-based method. Our analysis of the components of the sets of overlapping genes supports the utility of the PRSOG method.
All site content, except where otherwise noted, is licensed under a Creative Commons Attribution 3.0 License.