But beyond these individual methods for data processing, the profound biological insights of ENCODE undoubtedly come from computational approaches that integrated multiple data types. For example, by combining data on DNA methylation, DNA accessibility and transcription-factor expression. Thurman et al. (4) provide fascinating insight into the causal role of DNA methylation in gene silencing. They find that transcription-factor binding sites are, on average, less frequently methylated in cell types that express those transcription factors, suggesting that binding-site methylation often results from a passive mechanism that methylates sites not bound by transcription factors.
Despite the extensive functional information provided by ENCODE, we are still far from the ultimate goal of understanding the function of the genome in every cell of every person, and across time within the same person. Even if the throughput rate of the ENCODE profiling methods increases dramatically, it is clear that brute-force measurement of this vast space is not feasible. Rather, we must move on from descriptive and correlative computational analyses, and work towards deriving quantitative models that integrate the relevant protein, RNA and chromatin components. We must then describe how these components interact with each other, how they bind the genome and how these binding events regulate transcription.
If successful, such models will be able to predict the genome’s function at times and in settings that have not been directly measured. By allowing us to determine which assumptions regarding the physical interactions of the system lead to models that better explain measured patterns, the ENCODE data provide an invaluable opportunity to address this next immense computational challenge.
