Limits...
Power and sample size estimation for epigenome-wide association scans to detect differential DNA methylation.

Tsai PC, Bell JT - Int J Epidemiol (2015)

Bottom Line: We performed simulations to estimate power under the case-control and discordant MZ twin EWAS study designs, under a range of epigenetic risk effect sizes and conditions.Our analyses highlighted several factors that significantly influenced EWAS power, including sample size, epigenetic risk effect size, the variance of DNA methylation at the locus of interest and the correlation in DNA methylation patterns within the twin sample.Our results can help guide EWAS experimental design and interpretation for future epigenetic studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Twin Research and Genetic Epidemiology, King's College London, London, UK.

No MeSH data available.


Power of case-control EWAS. Estimates are obtained for a range of sample sizes, using (A) mean difference and (B) methOR effects, at nominal (upper panel) and genome-wide (lower panel) significance thresholds. Each line represents the power curve under different case-control sample sizes from 10 to 500 pairs of cases and controls.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4588864&req=5

dyv041-F3: Power of case-control EWAS. Estimates are obtained for a range of sample sizes, using (A) mean difference and (B) methOR effects, at nominal (upper panel) and genome-wide (lower panel) significance thresholds. Each line represents the power curve under different case-control sample sizes from 10 to 500 pairs of cases and controls.

Mentions: Power simulations were performed under the case-control EWAS design, by sampling effect sizes based on the mean difference in DNA methylation between cases and controls. Cases were selected from one of eight case distributions and controls were drawn from the control distribution, using 1000 permutations per simulation. Simulations were performed with mean difference effects from 1% to 60% and with increasing sample sizes from 10 to 500 pairs of cases and controls, that is, 20 to 1000 individuals altogether (Figure 3A, Supplementary Table 1a, available as Supplementary data at IJE online). Figure 3A shows the mean difference required to achieve 80% power at different sample sizes at P-value thresholds of 0.05 (single locus threshold, upper plot) and 1 × 10−6 (genome-wide threshold, lower plot). For example, a sample size of 100 cases and 100 controls results in over 80% power to detect a 4.5% mean difference (mean methOR = 1.32) in methylation at nominal significance (P = 0.05). However, at a genome-wide significance (P = 1 × 10−6) the same sample size gives over 80% power to detect a much larger effect size of 11% mean difference (mean methOR = 1.81). The results of the Wilcoxon test are shown in Supplementary Table 1 and 2, available as Supplementary data at IJE online. We also performed power estimation under the one case-multiple controls scenario. We show results from one case:two controls and one case:four controls study design (Supplementary Table 1b and c, available as Supplementary data at IJE online) and, as expected, power increases when the sample size of the control group increases.Figure 3.


Power and sample size estimation for epigenome-wide association scans to detect differential DNA methylation.

Tsai PC, Bell JT - Int J Epidemiol (2015)

Power of case-control EWAS. Estimates are obtained for a range of sample sizes, using (A) mean difference and (B) methOR effects, at nominal (upper panel) and genome-wide (lower panel) significance thresholds. Each line represents the power curve under different case-control sample sizes from 10 to 500 pairs of cases and controls.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4588864&req=5

dyv041-F3: Power of case-control EWAS. Estimates are obtained for a range of sample sizes, using (A) mean difference and (B) methOR effects, at nominal (upper panel) and genome-wide (lower panel) significance thresholds. Each line represents the power curve under different case-control sample sizes from 10 to 500 pairs of cases and controls.
Mentions: Power simulations were performed under the case-control EWAS design, by sampling effect sizes based on the mean difference in DNA methylation between cases and controls. Cases were selected from one of eight case distributions and controls were drawn from the control distribution, using 1000 permutations per simulation. Simulations were performed with mean difference effects from 1% to 60% and with increasing sample sizes from 10 to 500 pairs of cases and controls, that is, 20 to 1000 individuals altogether (Figure 3A, Supplementary Table 1a, available as Supplementary data at IJE online). Figure 3A shows the mean difference required to achieve 80% power at different sample sizes at P-value thresholds of 0.05 (single locus threshold, upper plot) and 1 × 10−6 (genome-wide threshold, lower plot). For example, a sample size of 100 cases and 100 controls results in over 80% power to detect a 4.5% mean difference (mean methOR = 1.32) in methylation at nominal significance (P = 0.05). However, at a genome-wide significance (P = 1 × 10−6) the same sample size gives over 80% power to detect a much larger effect size of 11% mean difference (mean methOR = 1.81). The results of the Wilcoxon test are shown in Supplementary Table 1 and 2, available as Supplementary data at IJE online. We also performed power estimation under the one case-multiple controls scenario. We show results from one case:two controls and one case:four controls study design (Supplementary Table 1b and c, available as Supplementary data at IJE online) and, as expected, power increases when the sample size of the control group increases.Figure 3.

Bottom Line: We performed simulations to estimate power under the case-control and discordant MZ twin EWAS study designs, under a range of epigenetic risk effect sizes and conditions.Our analyses highlighted several factors that significantly influenced EWAS power, including sample size, epigenetic risk effect size, the variance of DNA methylation at the locus of interest and the correlation in DNA methylation patterns within the twin sample.Our results can help guide EWAS experimental design and interpretation for future epigenetic studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Twin Research and Genetic Epidemiology, King's College London, London, UK.

No MeSH data available.