TY - JOUR T1 - Correcting for cell-type heterogeneity in epigenome-wide association studies: premature analyses and conclusions JF - bioRxiv DO - 10.1101/121533 SP - 121533 AU - Shijie C Zheng AU - Stephan Beck AU - Andrew E. Jaffe AU - Devin C. Koestler AU - Kasper D. Hansen AU - Andres E. Houseman AU - Rafael A. Irizarry AU - Martin Widschwendter AU - Andrew E. Teschendorff Y1 - 2017/01/01 UR - http://biorxiv.org/content/early/2017/03/28/121533.abstract N2 - Recently, a study by Rahmani et al [1] claimed that a reference-free cell-type deconvolution method, called ReFACTor, leads to improved power and improved estimates of cell-type composition compared to competing reference-free and reference-based methods in the context of Epigenome-Wide Association Studies (EWAS). However, we identified many critical flaws (both conceptual and statistical in nature), which seriously question the validity of their claims. We outlined constructive criticism in a recent correspondence letter, Zheng et al [2]. The purpose of this letter is two-fold. First, to present additional analyses, which demonstrate that our original criticism is statistically sound. Second, to highlight additional serious concerns, which Rahmani et al have not yet addressed. In summary, we find that ReFACTor has not been demonstrated to outperform state-of-the-art reference-free methods such as SVA or RefFreeEWAS, nor state-of-the-art reference-based methods. Thus, the claim by Rahmani et al (a claim reiterated in their recent response letter [3]) that ReFACT or represents an advance over the state-of-the-art is not supported by an objective and rigorous statistical analysis of the data. ER -