Small RNA mediated targeting, primarily miRNA targeting, is a process by which RNA translation and stability are regulated. Recent developments in Next Generation Sequencing techniques, specifically the isolation of chimeric reads, has allowed us the identification of unambiguous miRNA:target duplexes from sequencing data. Deep Learning is a family of cutting edge machine learning techniques that have been applied to pattern recognition problems with great success. This project involves the development of a Deep Learning system for the identification of miRNA (or other small RNA) driver sequences, trained on chimeric read data. The student will work with experimental data from CLASH experiments, both previously published (Ago1-CLASH, Helwak et al) and currently under publication from our research group (Ago2-CLASH).
A central aspect of the project is the concept of highly imbalanced datasets. Briefly, in the context of small RNA binding, there are hundreds of negative putative binding sites for each positive binding site in natural contexts. This imbalance leads to low precision of prediction when models are trained on balanced datasets. However, training on realistic imbalanced datasets is computationally expensive. Previously, we developed a method termed Iterative Background Selection (Georgakilas et al) that aimed to solve this problem for Convolutional Neural Networks used in genomic annotation. The student will re-implement this method, and others proposed in bibliography, and perform a thorough comparison using the experimental CLASH datasets.
Particular goals of the work are:

• Develop Deep Learning system for small RNA identification based on chimeric reads, based on Convolutional Neural Network architecture
• Identify and evaluate training methods dealing with *highly imbalanced datasets*.
• Evaluate precision/sensitivity of prediction.
• Dissemination of method (publication)

Helwak A, Kudla G, Dudnakova T, Tollervey D. Mapping the human miRNA interactome by CLASH reveals frequent noncanonical binding. Cell. 2013;153(3):654-665. doi:10.1016/j.cell.2013.03.043
Georgakilas, G.K., Grioni, A., Liakos, K.G. et al. Multi-branch Convolutional Neural Network for Identification of Small Non-coding RNA genomic loci. Sci Rep 10, 9486 (2020). https://doi.org/10.1038/s41598-020-66454-3