Table 2.

Assumptions in the learning hierarchical-pair model are supported by biological knowledge.

Model assumptions	Biological findings	Regulation
Competition	A transcription factor chooses a binding locus among candidates, depending on the openness ratio of the chromatin	Epigenomic
Amplification	Transcriptional coactivators with histone acetyltransferase activity relax the chromatin structure
	Transcription opens the chromatin, and the open chromatin structure induces transcription
Bias (no extinction) Additive increase	Whole-genome in every somatic cell	Genetic
	Conventional genetic regulation of transcription
Error (approximated)-dependent decay	Cellular stress responses	Dependent on cell and environment Feedback from the current fitness
	Rough evaluation of the current state
	Histone deacetylases and DNA methyltransferases close the chromatin structure
	Non-coding RNA-dependent cleavage
	RNA-mediated epigenomic modification
Hierarchical-pair architecture	Signal transduction cascades for gene expression	Genetic
	Topologically associated domains (TADs)
Competitive amplification in hierarchical pairs	Active and expressed cascades are preferentially selected and activated	Cell-type dependent Post-translational
	Kinase is activated by phosphorylation at multiple sites
Error-dependent decay in hierarchical pairs	Cellular stress responses	Dependent on cell and environment
	Dephosphorylation
	Polyubiquitin dependent degradation
	RNA-mediated epigenomic modification

Epigenetic regulations, which are highly variable depending on cell type, can be interpreted as a process of competitive amplification. The decay rate is roughly regulated at several levels by the fitness of the current expression pattern in each pair. The correct expression level of each gene is not supervised in real cells. Instead, two functionally related gene-groups are regulated in a pair, in which the inappropriate expression ratio induces cellular stress, increases the decay, and destabilizes the ratio. As a possible feedback regulation for the error-dependent decay, cleaved mRNA fragments coding excessive proteins may close the genome loci. Hierarchical pairs are genetically determined and consistent in all cell-types.