Dr. Dongmei Li, Biostatistician and Associate Professor, and Dr. Timothy Dye, Director of CTSI Bioinformatics and Professor, led a team of scientists to evaluate the statistical methods used to analyze DNA methylation data. Their paper was recently published in BMC Bioinformatics and has been noted by the genomics and biostatistical fields. DNA methylation is a cellular process for altering the chemical structure of DNA and is used by cells to regulate gene expression. Hypermethylation of gene promotors has been associated with cancer and some cancer therapies reverse hypermethylation. The ability to perform large-scale examinations of DNA methylation through microarrays paired with robust statistical analyses will allow investigators to accurately interpret data and identify methylation patterns attributable to disease. However, there are many biostatistical tools to choose from for DNA methylation analysis and it is not known how each method performs. Drs. Li and Dye compared the false discovery rate, power and stability of six commonly used statistical approaches: Wilcoxon rank sum test, Kolmogorov-Smirnov test, permutation test, empirical Bayes method, t-test, and bump hunting method. They used two publically available data sets to validate their results, the United Kingdom Ovarian Cancer Population Study dataset and a rheumatoid arthritis dataset. They found that all biostatistical tests performed equally when using medium or large samples sizes (12-24 samples). However, in small data sets (3-6 samples) the bump hunting method and empirical Bayes method performed better.
“This work has received a lot of attention because the use of DNA methylation analysis for investigating disease states in the genome is rapidly expanding” says Li.
Citation: Li D, Xie Z, Pape ML, Dye T. An evaluation of statistical methods for DNA methylation microarray data analysis. BMC Bioinformatics. 2015 Jul 10;16:217. doi: 10.1186/s12859-015-0641-x. PubMed PMID: 26156501; PubMed Central PMCID: PMC4497424.