Leveraging large-scale datasets to understand the interaction between the genome and the epigenome

Epigenetics is typically described as a layer of molecular information above and beyond the DNA sequence. While this conceptualization is certainly accurate to some extent, there is also a tight connection between the genome and the epigenome, as the basic components of the epigenetic machinery (EM) are DNA-encoded. This thesis focuses on four such genetic components: genes encoding for the proteins of the histone machinery, genes encoding for the proteins of the DNA methylation machinery, genes encoding for chromatin remodelers, and CpG dinucleotides. We first perform a systematic analysis of all human EM genes, and characterize them with respect to their tolerance to variation, both at the whole-gene level, and the local, protein domain level. We then discover a systems-level property (co-expression), that is specifically exhibited by a large subset of variation-intolerant EM genes, and may be particularly relevant to their involvement in neurodevelopment. Finally, we shift our focus on the CpG dinucleotides. We show that a high promoter CpG density is not merely a generic feature of human promoters, but is preferentially encountered at the promoters of the most loss-of-function intolerant genes. This coupling calls into question the prevailing view that CpG islands are not subject to selection. It also has practical utility, as it allows us to train a simple and easily interpretable predictive model of loss-of-function intolerance that outperforms existing predictors and classifies 1,760 genes - which are currently unascertained - as highly loss-of-function-intolerant or not. Together, the results presented in this thesis provide new insights into the interaction between the genome and the epigenome