Predicting Active Enhancers From Chromatin Accessibility and RNA-seq Datasets

Genomic annotation tools transform vast amounts of genomic data into interpretable labels, and help identify regulatory elements. As genomic assays improve to probe smaller samples of fewer cells, they become capable of characterizing new cell types. Genomic annotation methods must adapt to make sense of this wealth of new information despite the relative dearth of data on protein-DNA interactions on these new cell types, especially for use in locating regulatory elements. This project investigates the effectiveness of Segway, a genomic annotation tool, on minimal amounts of information. This project aims to improve upon existing methods by investigating two new supervised approaches to predict specific regulatory regions (enhancers), elucidating their location given only chromatin accessibility and RNA-seq data. The results show that out of the algorithms available from scikit-learn, an extremely randomized trees classifier is most capable of locating enhancers using minimal data (auPR 0.65).M.Sc