Machine-learning to create predictive models of genetic regulation

The project follows our recent papers on the use of penalised (regularized) regression models to form gene specific regulatory models using large data sets of transcription factor binding, epigenetics and chromatin structure. We will take advantage of the increasing availability of 3D chromosome interaction data to train and test our methods, and we will extend our methodology to machine learning methods with which we have previously published experience (e.g. support vector machines, neural networks).

The project will directly employ machine learning methods using data from public sources (the ENCODE/modENCODE projects and others) as well as locally generated to data to form predictive regulatory models for genes. It aligns to the BBSRC remit of fundamental biological understanding, and will contribute to the interpretation of other large genomic data sets, for instance through understanding regulatory SNPs derived from GWAS studies in man and other species.