Good Leaver

Identification of molecular markers of platinum sensitivity in ovarian and breast cancers using deep learning 

Posted 2 years ago


The Institute of Cancer Research London , is one of the world’s most influential cancer research organisations, with an outstanding record of achievement dating back more than 100 years. The Institute of Cancer Research (ICR) discovers more new cancer drugs than any other academic institution globally, and with The Royal Marsden runs one of the world’s largest phase I clinical trial units for cancer. It is also a world leader in cancer biology and genetics, and in developing new forms of high-precision radiotherapy. Its mission is to make the discoveries that will defeat cancer.   

Collaborators:  Dr. Syed Haider, Prof. Andrew Tutt  

Location:  Remote   

Project Requirements   

  • Experience applying unsupervised machine learning to large data sets   
  • Experience with AutoEncoder Neural Networks desired 
  • Understanding of logistic regression 
  • PhD/MSc in STEM discipline   
  • Programming skills in R or Python  
  • Bioinformatics knowledge useful but not essential as training will be provided 

Project Summary   

Ovarian cancer is the eighth most common cancer amongst women worldwide, with around 300,000 new cases reported in 2018 [1]. There were almost 185,000 ovarian cancer deaths reported in the same period globally [1]. Management of ovarian cancer involves surgery and platinum–taxane chemotherapy [2].  After treatment, platinum-resistant cancer recurs within six months in around 25% of patients. Patients with no recurrence in six months are considered platinum-sensitive. The strongest marker of this sensitivity remains mutations in BRCA1/2 genes and deficiencies in the homologous recombination (HR) repair pathway [3]. However, some BRCA1/2 deficient cancers frequently acquire resistance. Conversely, many BRCA1/2 efficient cancers exhibit platinum-sensitivity. These challenges in the management of ovarian cancer are also presented by the triple negative breast cancers (TNBC) [4] where the current recommendations for the use of platinum therapy outside clinical trial setting could be further refined [5]. Therefore, investigating platinum sensitivity as a collective phenomenon may reveal new insights into molecular mechanisms of sensitivity and identification of patients suitable for platinum or targeted therapies, such as PARP inhibitors.    

To delineate molecular characteristics of platinum sensitive cancers, we propose to take a machine learning approach. Machine learning has revolutionised the discovery of complex association between molecular features and clinical outcome [6,7]. In particular, Autoencoder neural networks, an unsupervised machine learning algorithm, have shown promising results in denoising gene expression profiles [8].   

In this study, we will test if Autoencoder neural networks are able to identify latent variables that can stratify platinum sensitive ovarian cancers and TNBCs into BRCA1/2 mutant (and HR-deficient) and BRCA1/2 efficient groups. Subsequently, interpretation of predictive latent variables will be performed by mapping these back to molecular pathways, thereby highlighting genes and pathways associated with platinum-sensitivity. The study will take an integrative approach by using genome-wide DNA and RNA profiles of ovarian cancers and TNBCs from TCGA project [2,9].   


[1] Bray, F. et al. (2018) Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin.  

[2] The Cancer Genome Atlas Research Network (2011) Integrated genomic analyses of ovarian carcinoma. Nature  

[3] Jiang, X. et al. (2019) PARP inhibitors in ovarian cancer: Sensitivity prediction and resistance mechanisms. J Cell Mol Med.  

[4] Tutt, A, et al. (2018) Carboplatin in BRCA1/2-mutated and triple-negative breast cancer BRCAness subgroups: the TNT Trial. Nature Medicine 

[5] Goetz, M. et al. (2018) NCCN Guidelines Insights: Breast Cancer, Version 3.2018. JNCCN J Natl Compr. Cancer Network 

[6] Kourou, K. et al. (2015) Machine learning applications in cancer prognosis and prediction. Computational and Structural Biotechnology Journal  

[7] Kawakami, E. et al. (2019) Application of artificial intelligence for preoperative diagnostic and prognostic prediction in epithelial ovarian cancer based on blood biomarkers. Clinical Cancer Research  

[8] Xie, R. et al. (2016) A deep auto-encoder model for gene expression prediction. BMC Genomics  

[9] The Cancer Genome Atlas Research Network (2012) Comprehensive molecular portraits of human breast tumours. Nature  

Apply Online

A valid email address is required.
A valid phone number is required.
Copy link
Powered by Social Snap