Using machine learning to predict cell-type specific effects of genetic variants which influence genome regulation.

This PhD project is focused on using machine learning techniques to develop novel classifiers for predicting how changes in DNA sequences alter genomic regulatory features. Many regulatory proteins recognise particular DNA sequences known as motifs, for instance, EcoRI only binds to GAATTC. DNA sequences can be converted into a machine interpretable format, using one-hot encoding. The candidate will use publicly available and inhouse datasets of genomic regulatory features to train models. Machine learning techniques will be used to predict the cell-type specific regulatory effects of genetic variants. We will provide several true-positive datasets, wherein the effect of genetic mutations on particular regulatory features has been measured. These will form validation datasets to evaluate how well the trained classifier works. We are interested in how improvements in the machine learning approach (e.g. use of transfer learning, recurrent attentional networks or graph convolution networks) can be used to improve upon existing methods. The candidate will use these techniques to identify causal pathways and candidate drug targets for neurodegenerative diseases.