Predicting transcription from chromatin interactions using machine learning

Abstract

Over the past decade, developments in the field of functional genomics have spurred improvements in high-throughput sequencing technologies. This has revolutionised the way in which gene regulation is studied. One novel approach would be the incorporation of machine learning to predict gene expression from epigenetic components. Recently, it has been illustrated that chromatin interactions play crucial roles in regulating gene expression. While there are various computational methods in predicting transcription from functional genomics data, chromatin interactions have not been utilised in such algorithms due to the scarcity of genome-wide chromatin interactions data. Here, we developed a machine learning model to predict transcription from both epigenetic factors and chromatin interactions. We found that chromatin interactions are important features in predicting transcription, with reasonable accuracy and good across-sample performances. Our model performance was validated by bioinformatics analyses in CTCF-depleted cells, where we found an enrichment in genes associated with transcriptional regulation and splicing. Our study further supports the functional links between chromatin interactions and gene regulation. Interestingly, we also discovered a novel insight in the role of chromatin interactions in splicing. This hopefully paves the way for future research to better unravel the interplay between splicing and chromatin interactions.