A chromatin structure‐based model accurately predicts DNA replication timing in human cells

The metazoan genome is replicated in precise cell lineage‐specific temporal order. However, the mechanism controlling this orchestrated process is poorly understood as no molecular mechanisms have been identified that actively regulate the firing sequence of genome replication. Here, we develop a mechanistic model of genome replication capable of predicting, with accuracy rivaling experimental repeats, observed empirical replication timing program in humans. In our model, replication is initiated in an uncoordinated (time‐stochastic) manner at well‐defined sites. The model contains, in addition to the choice of the genomic landmark that localizes initiation, only a single adjustable parameter of direct biological relevance: the number of replication forks. We find that DNase‐hypersensitive sites are optimal and independent determinants of DNA replication initiation. We demonstrate that the DNA replication timing program in human cells is a robust emergent phenomenon that, by its very nature, does not require a regulatory mechanism determining a proper replication initiation firing sequence. Synopsis A mechanistic model predicts cell lineage‐specific DNA replication timing based on the location of DNase‐hypersensitivity data alone. With essentially no parameters to adjust for different cell types, the model is truly predictive even for cells where timing data are not available. The mechanistic model predicts replication timing in human cells with an accuracy approaching the limit set by experimental noise. In the model, the timing program results from purely time‐stochastic initiation at well‐localized initiation sites and it is determined by the location of initiation sites alone regardless of initiation probabilities. Replication initiation sites are optimally localized by DNase‐hypersensitive sites. Graphical Abstract A mechanistic model predicts cell lineage‐specific DNA replication timing based on the location of DNase‐hypersensitivity data alone. With essentially no parameters to adjust for different cell types, the model is truly predictive even for cells where timing data are not available.