Abstract
DNA methylation, a universal epigenetic mechanism, is pivotal in regulating transcription andsuppressing gene expression in various ways, and its interference is associated with numerous
complex diseases. Graphical models or networks illustrate the statistical dependence among
multiple variables and are widely used in biology, such as gene regulatory networks.
In this study, we explored and compared two causal network inference applications to an-
alyze the relationship between DNA methylation and transcription. To achieve this, we gener-
alized the MRTrios package to handle different cancer types, aiming to gain insights into
the underlying mechanisms of gene regulation. Our analysis involved studying the relation-
ships between transcription and methylation in these cancer types using data from The Cancer
Genome Atlas (TCGA) consortium and the Genomics Data Common portal (GDC). The for-
mulated trios consist of the Copy Number Alteration (CNA) of a gene, the expression (E) of
tha gene, and the methylation (M) of a site located nearby or within the same gene.
We then applied MRGN, a novel causal network inference method that considers many
confounding variables under the principle of Mendelian randomization. Using the Bayesian
Inference, we calculated the posterior probabilities of edges in the inferred
models using genetic variants and the identified confounders in trios for one of the cancer
types. Comparing the results of the causal networks obtained from the Machine Learning method
and Bayesian Learning method, we observe that most of the causal models generated for each trio are
similar for both methods with minor differences.
Our comparative analysis highlights the strengths and limitations of each causal network
inference method in underlying the complex mechanisms of DNA methylation and Transcrip-
tion. Our findings provide an advancing understand to researchers on selecting appropriate
inference methodologies for analyzing regulatory networks.