Limits...
M3D: a kernel-based test for spatially correlated changes in methylation profiles.

Mayo TR, Schweikert G, Sanguinetti G - Bioinformatics (2014)

Bottom Line: We propose a non-parametric, kernel-based method, M(3)D, to detect higher order changes in methylation profiles, such as shape, across pre-defined regions.The test statistic explicitly accounts for differences in coverage levels between samples, thus handling in a principled way a major confounder in the analysis of methylation data.Empirical tests on real and simulated datasets show an increased power compared to established methods, as well as considerable robustness with respect to coverage and replication levels.

View Article: PubMed Central - PubMed

Affiliation: IANC, School of Informatics, University of Edinburgh, Edinburgh EH8 9AB and Wellcome Trust Centre for Cell Biology, University of Edinburgh, Edinburgh EH9 3JR, UK.

Show MeSH
Simulation results. (a–c) We plot here the coverage MMD against the full MMD metric for all methods. The M3D test statistic is their difference, the distance in the x axis from the diagonal line. Each point is a CpG cluster, with black points being unchanged. DMRs are shaded according to whether they are called (M3D) or missed (BSmooth). (a) M3D identifies a clear relationship and calls almost all the clusters. (b) BSmooth calls some of the clusters but makes both types of error (Table 1). Classification bears little resemblance to the M3D method. (c) MAGI calls fewer regions, again with little semblance to the M3D method. (d) Histogram of test statistics for replicate values (blue) and with simulated changes (red), log scale
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4380032&req=5

btu749-F2: Simulation results. (a–c) We plot here the coverage MMD against the full MMD metric for all methods. The M3D test statistic is their difference, the distance in the x axis from the diagonal line. Each point is a CpG cluster, with black points being unchanged. DMRs are shaded according to whether they are called (M3D) or missed (BSmooth). (a) M3D identifies a clear relationship and calls almost all the clusters. (b) BSmooth calls some of the clusters but makes both types of error (Table 1). Classification bears little resemblance to the M3D method. (c) MAGI calls fewer regions, again with little semblance to the M3D method. (d) Histogram of test statistics for replicate values (blue) and with simulated changes (red), log scale

Mentions: The M3D statistic will therefore be different from zero when there is a change in the methylation profile, independently of a change in the coverage profile. As a consequence, M3D between replicate RRBS experiments (which do not necessarily have identical coverage) should be close to zero or, equivalently, the full MMD should be equal to the coverage MMD. This is borne out in the data; the metrics strongly agree over replicates. Testing equality of metrics over 102 ENCODE RRBS datasets gives an R2 of 0.95. This can be seen in Supplementary Figure 2; specific examples can also be seen in Figures 2(a–c) and 4(a–c), where the dense region around the diagonal represents unchanged DMRs with M3D close to zero.Fig. 2.


M3D: a kernel-based test for spatially correlated changes in methylation profiles.

Mayo TR, Schweikert G, Sanguinetti G - Bioinformatics (2014)

Simulation results. (a–c) We plot here the coverage MMD against the full MMD metric for all methods. The M3D test statistic is their difference, the distance in the x axis from the diagonal line. Each point is a CpG cluster, with black points being unchanged. DMRs are shaded according to whether they are called (M3D) or missed (BSmooth). (a) M3D identifies a clear relationship and calls almost all the clusters. (b) BSmooth calls some of the clusters but makes both types of error (Table 1). Classification bears little resemblance to the M3D method. (c) MAGI calls fewer regions, again with little semblance to the M3D method. (d) Histogram of test statistics for replicate values (blue) and with simulated changes (red), log scale
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4380032&req=5

btu749-F2: Simulation results. (a–c) We plot here the coverage MMD against the full MMD metric for all methods. The M3D test statistic is their difference, the distance in the x axis from the diagonal line. Each point is a CpG cluster, with black points being unchanged. DMRs are shaded according to whether they are called (M3D) or missed (BSmooth). (a) M3D identifies a clear relationship and calls almost all the clusters. (b) BSmooth calls some of the clusters but makes both types of error (Table 1). Classification bears little resemblance to the M3D method. (c) MAGI calls fewer regions, again with little semblance to the M3D method. (d) Histogram of test statistics for replicate values (blue) and with simulated changes (red), log scale
Mentions: The M3D statistic will therefore be different from zero when there is a change in the methylation profile, independently of a change in the coverage profile. As a consequence, M3D between replicate RRBS experiments (which do not necessarily have identical coverage) should be close to zero or, equivalently, the full MMD should be equal to the coverage MMD. This is borne out in the data; the metrics strongly agree over replicates. Testing equality of metrics over 102 ENCODE RRBS datasets gives an R2 of 0.95. This can be seen in Supplementary Figure 2; specific examples can also be seen in Figures 2(a–c) and 4(a–c), where the dense region around the diagonal represents unchanged DMRs with M3D close to zero.Fig. 2.

Bottom Line: We propose a non-parametric, kernel-based method, M(3)D, to detect higher order changes in methylation profiles, such as shape, across pre-defined regions.The test statistic explicitly accounts for differences in coverage levels between samples, thus handling in a principled way a major confounder in the analysis of methylation data.Empirical tests on real and simulated datasets show an increased power compared to established methods, as well as considerable robustness with respect to coverage and replication levels.

View Article: PubMed Central - PubMed

Affiliation: IANC, School of Informatics, University of Edinburgh, Edinburgh EH8 9AB and Wellcome Trust Centre for Cell Biology, University of Edinburgh, Edinburgh EH9 3JR, UK.

Show MeSH