Targeted, Robust, and Model-free Differential Methylation Analysis
Authors: Nima Hejazi and Mark van der Laan
methyvim is an R package that provides facilities for differential
methylation analysis based on variable importance measures (VIMs),
statistical target parameters inspired by causal inference.
The statistical methodology implemented computes targeted minimum loss estimates of several well-characterized variable importance measures:
For discrete-valued treatments or exposures:
-
The average treatment effect (ATE): The effect of a binary exposure or treatment on the observed methylation at a target CpG site is estimated, controlling for the observed methylation at all other CpG sites in the same neighborhood as the target site, based on an additive form. In particular, the parameter estimate represents the additive difference in methylation that would have been observed at the target site had all observations received the treatment versus the scenario in which none received the treatment.
-
The relative risk (RR): The effect of a binary exposure or treatment on the observed methylation at a target CpG site is estimated, controlling for the observed methylation at all other CpG sites in the same neighborhood as the target site, based on an geometric form. In particular, the parameter estimate represents the multiplicative difference in methylation that would have been observed at the target site had all observations received the treatment versus the scenario in which none received the treatment.
For continuous-valued treatments or exposures (WIP: support planned):
- A nonparametric variable importance measure (NPVI) (Chambaz, Neuvial, and van der Laan 2012): The effect of continous-valued exposure or treatment (the observed methylation at a target CpG site) on an outcome of interest is estimated, controlling for the observed methylation at all other CpG sites in the same neighborhood as the target (treatment) site, based on a parameter that compares values of the treatment against a reference value taken to be the null. In particular, the implementation provided is designed to assess the effect of differential methylation at the target CpG site on a (typically) phenotype-level outcome of interest (e.g., survival), in effect providing an nonparametric evaluation of the impact of methylation at the target site on said outcome.
These methods allow differential methylation effects to be quantified in a manner that is largely assumption-free, especially of the variety exploited in parametric models. The statistical algorithm consists in several major steps:
- Pre-screening of genomic sites is used to isolate a subset of sites
for which there is cursory evidence of differential methylation. For
the sake of computational feasibility, targeted minimum loss-based
estimates of VIMs are computed only for this subset of sites.
Several screening approaches are available, adapting core routines
from the following R packages:
limma,tmle.npvi. - Nonparametric VIMs are estimated for the specified parameter,
currently adapting routines from the
tmle.npviandtmleR packages. - Since pre-screening is performed prior to estimating VIMs, we make use of a multiple testing correction uniquely suited to such settings. Due to the multiple testing nature of the estimation problem, a variant of the Benjamini & Hochberg procedure for controlling the False Discovery Rate (FDR) is applied (Benjamini and Hochberg 1995). Specifically, we apply the modified marginal Benjamini & Hochberg step-up False Discovery Rate controlling procedure for multi-stage analyses (FDR-MSA) (Tuglus and van der Laan 2009).
For a general discussion of the framework of targeted minimum loss estimation and its myriad applications, the canonical references are van der Laan and Rose (2011) and van der Laan and Rose (2018). Hernan and Robins (2019) and Pearl (2000) may be of interest to those desiring a more general introduction to statistical causal inference.
For standard use, install from
Bioconductor using
BiocManager:
if (!requireNamespace("BiocManager", quietly=TRUE)) {
install.packages("BiocManager")
}
BiocManager::install("methyvim")To contribute, install the bleeding-edge development version from
GitHub via remotes:
remotes::install_github("nhejazi/methyvim")Current and prior Bioconductor releases are available under branches with numbers prefixed by “RELEASE_”. For example, to install the version of this package available via Bioconductor 3.6, use
remotes::install_github("nhejazi/methyvim", ref = "RELEASE_3_6")For details on how to best use the methyvim R package, please consult
the most recent package
vignette
available through the Bioconductor
project.
If you encounter any bugs or have any specific feature requests, please file an issue.
Contributions are very welcome. Interested contributors should consult our contribution guidelines prior to submitting a pull request.
After using the methyvim R package, please cite the following:
@article{hejazi2018methyvim,
doi = {10.12688/f1000research.16047.1},
url = {https://dx.doi.org/10.12688/f1000research.16047.1},
year = {2018},
publisher = {Faculty of 1000 Ltd},
volume = {7},
number = {1424},
author = {Hejazi, Nima S and Phillips, Rachael V and Hubbard, Alan E
and {van der Laan}, Mark J},
title = {{methyvim}: Targeted, robust, and model-free differential
methylation analysis in {R}},
journal = {F1000Research}
}
@manual{hejazi2019methyvimbioc,
author = {Hejazi, Nima S and {van der Laan}, Mark J},
title = {{methyvim}: Targeted, robust, and model-free differential
methylation analysis},
doi = {10.18129/B9.bioc.methyvim},
url = {https://bioconductor.org/packages/methyvim},
note = {R package version 1.8.0}
}
- R/
methyvimData- R package with sample experimental DNA methylation data for use as an example with this analysis package.
The development of this software was supported in part through grants from the National Institutes of Health: T32 LM012417-02, R01 ES021369-05, and P42 ES004705-29.
© 2017-2019 Nima S. Hejazi
The contents of this repository are distributed under the MIT license.
See file LICENSE for details.
Benjamini, Yoav, and Yosef Hochberg. 1995. “Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing.” Journal of the Royal Statistical Society. Series B (Methodological). JSTOR, 289–300.
Chambaz, Antoine, Pierre Neuvial, and Mark J van der Laan. 2012. “Estimation of a Non-Parametric Variable Importance Measure of a Continuous Exposure.” Electronic Journal of Statistics 6. NIH Public Access: 1059.
Hernan, Miguel A, and James M Robins. 2019. Causal Inference. Chapman & Hall / CRC Texts in Statistical Science. Taylor & Francis.
Pearl, Judea. 2000. Causality: Models, Reasoning, and Inference. Cambridge University Press.
Tuglus, Catherine, and Mark J van der Laan. 2009. “Modified FDR Controlling Procedure for Multi-Stage Analyses.” Statistical Applications in Genetics and Molecular Biology 8 (1). Walter de Gruyter: 1–15. https://doi.org/10.2202/1544-6115.1397.
van der Laan, Mark J, and Sherri Rose. 2011. Targeted Learning: Causal Inference for Observational and Experimental Data. Springer Science & Business Media.
———. 2018. Targeted Learning in Data Science: Causal Inference for Complex Longitudinal Studies. Springer Science & Business Media.