Understanding the rules of life

Bioscience for an integrated understanding of health

Category: Standard Studentships

Using AI and big data to identify a set of biologically validated drug targets for hard-to-treat cancers

Project No.2245

Primary Supervisor

Dr Frances Pearl – University of Sussex


Prof Michelle Garrett – University of Kent



Background The ultimate goal in cancer treatment is to identify the therapeutic vulnerabilities of a patient’s tumour and use this to design a personalised medicine regime.

The cost reduction of genomic technologies in the last few years, has allowed extensive genomic analysis of clinical samples but for most tumour types, we currently lack the ability to translate these data into a successful therapeutic

strategy. The Pearl bioinformatics laboratory have therefore developed a suite of artificial intelligence (AI) algorithms that use cancer genomic and other ‘big’ data sets to predict druggable vulnerabilities in cancer cells. The SLant and MexDrugs algorithms predict tumour-specific therapeutic strategies by identifying targetable synthetically lethal gene pairs of tumour suppressors through protein-protein interaction (PPI) data (SLant) or methylation data (MexDrugs). The DependANT algorithm uses mutation and expression data from cancer cell lines to modulate PPI networks to identify the genes that the cell has become dependent on. The accuracy of all these algorithms depends on the availability and quality of large datasets, which are increasing in size exponentially. Aims and Objectives The aim of the project is to apply these AI approaches to find novel treatments for hard to treat cancers including triple negative breast cancer, KRAS-mutated pancreatic cancer and oesophageal cancer. The project will involve (i) Identification and manipulation of multi-platform tumour and cancer cell-line data (Mutational, expression, methylation. CRISPR-CAS9), PPI data and drug and clinical trial data. (ii) Improvement of the accuracy of the DependANT algorithm using deep learning approaches and applying this to data from clinical cancer samples rather than culture-adapted cell lines. (iii) Application of the DependANT+ algorithm to find novel drug targets in hard-to-treat cancers (outlined above). (iv) Experimental validation of the predictions in the Garrett laboratory. Project output A set of novel, biologically validated drug targets for hard-to-treat cancers.