Applying large scale metanalysis of transcriptomic data to uncover hyper-responsive genes and prediction via machine learning

Abstract

With the increasing adoption of high-throughput transcriptomic technologies there has been efforts to leverage these pre-existing datasets to improve the biological interpretability of bulk transcriptomic data analyses. Researchers have observed that genes exhibit markedly different responsiveness to perturbation- i.e. a subset of genes are more likely to exhibit large changes to their expression. This thesis expands on this area by developing a novel approach to bulk transcriptomic data analysis which leverages pre-existing datasets. Additionally, this thesis also showcases novel approach to feature selection for scRNA-seq datasets which also leverages pre-existing transcriptomic data to improve the clustering of annotated scRNA-seq datasets. Finally, this thesis utilises machine learning models to identify the key genomic and transcript-based features of these genes that explain the differences in genes’ responsiveness to perturbation.

More information

Identifiers

ORCID for Brandon Coke: orcid.org/0000-0002-0847-6885

ORCID for Rob Ewing: orcid.org/0000-0001-6510-4001

Catalogue record

Export record