**PLEASE NOTE – the deadline for requesting a funding pack from Darwin Trust has now passed and completed funding applications must be submitted to Darwin Trust by 19th January. We can still accept applications for this project from self-funding students.
The classical model of cancer formation suggests that tumors arise from the stochastic accumulation of somatic mutations in key genes, called cancer driver genes, which equip an aberrant cell with the ability to escape apoptosis, elude the immune system and grow uncontrollably throughout the body [1].
However, high-throughput sequencing studies of many tissue types have recently found driver mutations in cancer genes also in normal cells, suggesting that somatic mutations are required but not sufficient to trigger tumorigenesis. It is then reasonable to hypothesise that mutations in non-driver genes are also required to transform a normal cell into malignant one; thus, tumor formation can be seen as a gene-network de-regulation process, rather than a single-gene mediated process. Importantly, cancer cells evolve and adapt during their lifespan to get proliferative advantage, but it is largely unknown when a cancer-related pathways get deregulated during this process; ultimately, this is a key aspect to understand disease progression, response to therapy and relapse.
While high-throughput omic technologies allow us to study the mutational and transcriptional landscape of cancer at gene-level, they cannot capture the underlying interactions underpinning tumorigenesis. However, protein-protein interaction data are becoming rapidly available, and integrating this information with omic experiments provides unique opportunities to identify cancer pathways and understand how they are activated or deactivated during tumor evolution.
Aims: The project aims at developing new graph neural networks to identify networks of genes associated with cancer and how they functionally evolve over time. Dr Stracquadanio’s group has recently developed deep learning methods to simultaneously identify cancer driver genes and pathways by integrating multi-omic data [2], along with statistical frameworks for characterising network properties of genes associated with cancer [3]. In collaboration with Dr Simpson, we will extend these methods to reconstruct pathways evolution from single cell RNA-seq data by introducing the concept of pseudo-timing to the graph neural network framework.
Training outcomes: The student will receive training in deep learning and graph neural networks, along with the methods to analyse big functional genomic datasets. The student will also learn how to write reproducible scientific pipelines and research software. We expect the student to build a competitive profile in machine learning and cancer biology, which ultimately will support a career in academia or industry.
The ideal candidate has a background in mathematics, synthetic biology, statistics, physics or related fields. He/she is strongly motivated to develop a competitive profile at the intersection of cancer biology, machine learning and genomics, while working in a fast-paced environment.
Stracquadanio lab: https://www.stracquadaniolab.org
Dr Stracquadanio’s Twitter: @DrStracquadanio
The School of Biological Sciences is committed to Equality & Diversity: https://www.ed.ac.uk/biology/equality-and-diversity