Learning the rules establishing the DNA replication landscape – RepliLand

DNA replication is a fundamental process of the cellular cycle. In metazoan replication starts stochastically at multiple sites called origin of replication. The factors responsible for selecting (i) the location of potential origins by origin licensing in the G1 phase and (ii) which potential origins to activate during S phase, remain largely unknown. These two steps combined together define the potential initiation landscape (PIL). Current experimental techniques target potential origin location by the mapping of origin components such as ORC and MCM proteins. However this does not capture the variable activation likelihood of potential origins. Another category of techniques probes the fulfilled replication program by determining activated origin locations, the mean replication timing (MRT) profiles or the replication fork directionality (RFD) profiles which contain information about the PIL but blurred by the origin passivation mechanism. The goal of the RepliLand project is to elucidate the complex combination of factors that govern PIL in metazoans and, exploiting this knowledge, develop an AI tool that will have the ability to predict the PIL from limited experimental data with the potential societal implication on human health.

We recently developed a method to uncover PIL from MRT and RFD data. However, these data are jointly available only on a very limited number of cell lines, mainly in human. We propose to improve the resolution of this method and to apply it to all cell lines where data is available in human and other metazoan species. Several works have shown a number of correlations between epigenetics marks and the regulation of replication initiation. We propose to develop an artificial intelligence (AI) method based on neural network to infer PIL in human, using the wide amount of epigenetic data available. We will then extent its domain of application using imputation techniques to mouse where three potential initiation landscapes and MRT profiles will be used to validate the method. To generalize further this imputation approach to other species or cell lines where the available epigenetic data are scarce, we will study its generalizability properties from a limited amount of information. If the developed AI tool proves to be robust to predict PIL in this setting, it could be applied to study the misregulation of replication during cancer progression, where typical studies are only able to record a limited amount of data. Finally, the project aims at deciphering the code governing the regulation of the replication initiation in human and other metazoans. Deploying explainable AI techniques on the potential initiation landscape prediction tool, we will question the synergistic or antagonistic contributions of chromatin context, transcription and genomic context either directly in cis or in trans or through indirect effects.