Predicting gene expression from epigenetics data
Yun Chen, Mette Jörgensen, Raivo Kolde and coworkers have showed the feasibility of predicting recruitment, elongation of stalling of RNA polymerase II, using machine learning methods and epigenetic data. The study also showed that a distinct set of epigenetic signals are necessary for recruitment of RNA polymerase II, but not elongation. The report was published in BMC Genomics (open access). Abstract is below:
Initiation and elongation of RNA polymerase II (RNAPII) transcription is regulated by both DNA sequence and chromatin signals. Recent breakthroughs make it possible to measure the chromatin state and activity of core promoters genome-wide, but dedicated computational strategies are needed to progress from descriptive annotation of data to quantitative, predictive models.
Here, we describe a computational framework which with high accuracy can predict the locations of core promoters, the amount of recruited RNAPII at the promoter, the amount of elongating RNAPII in the gene body, the mRNA production originating from the promoter and finally also the stalling characteristics of RNAPII by considering both quantitative and spatial features of histone modifications around the transcription start site (TSS). As the model framework can also pinpoint the signals that are the most influential for prediction, it can be used to infer underlying regulatory biology. For example, we show that the H3K4 di- and tri- methylation signals are strongly predictive for promoter location while the acetylation marks H3K9 and H3K27 are highly important in estimating the promoter usage. All of these four marks are found to be necessary for recruitment of RNAPII but not sufficient for the elongation. We also show that the spatial distributions of histone marks are almost as predictive as the signal strength and that a set of histone marks immediately downstream of the TSS is highly predictive of RNAPII stalling.
In this study we introduce a general framework to accurately predict the level of RNAPII recruitment, elongation, stalling and mRNA expression from chromatin signals. The versatility of the method also makes it ideally suited to investigate other genomic data