Precision medicine: how machine learning could change the way we treat cancer
by Alice Piller
Precision medicine is a hot topic amongst the scientific community seeing numerous researchers and practitioners swarm to the field to be at the forefront of this revolutionary concept. Precision medicine is going to change the way we treat diseases by enabling better prevention strategies, hastier diagnoses, quicker recovery, high survival rates, and fewer adverse side-effects. Sounds like a pretty sweet deal worth all the swarming, right? Well, unfortunately precision medicine is not quite a fully-fledged reality yet, but a study on pancreatic cancer  conducted by researchers at the University of Cape Town validifies the pursuit of this holy nectareous grail.
While not utilized at its full potential, precision medicine has been put into practice for decades. For example, a person’s blood type is taken into account before receiving a blood transfusion. On a grander scale, precision medicine would see an individual’s genes, environment and lifestyle being considered before administering a treatment. In a few cancer types, genetic markers exist that indicate the cancer’s sensitivity to certain drugs. For example, The KRAS mutation in colon cancer means it will likely respond well to a tyrosine kinase inhibitor drug, and a mutation in ABL1 in chronic myelogenous leukaemia means the cancer is likely going to be resistance to imatinib. 
Cancer is currently the second leading cause of death in the world with over 10 million cancer-related deaths occurring in 2019  and a projected 16.4 million deaths in 2040.  Current methods for outcome predictions and treatment decisions are largely based on tissues of origin and histological subtyping, leaving vast amounts of informative data unexplored. The researchers at the University of Cape Town lead by Musalula Sinkala, set out to characterise subtypes of pancreatic cancer based on molecular data from cell lines, which could highlight key biological pathways in driving oncogenesis and that could act as potential drug targets.
Sinkala et al. integrated proteomic, transcriptomic, DNA methylation, and miRNA data and identified two distinct pancreatic cancer subtypes, subtype-1 and subtype-2, using a machine learning clustering algorithm. Analysing the attributes of each subtype revealed several critical implications for clinical outcomes, treatment decisions, and drug targets.
Subtype-1 was found to be less aggressive than subtype-2 with a 75% survival rate versus only 35%. A possible reason for this is the increased DNA methylation in genes acting in key pathways including actin cytoskeleton regulation and focal adhesion.
Altered pathways can give great insight into drivers of oncogenesis and potential drug targets. Subtype-1 had hyperactivation and increased protein phosphorylation in the mTOR signaling pathway and displayed evidence of elevated ion channel and secretion pathway activities, whereas subtype-2 displayed hyperactivation and increased protein phosphorylation in cell cycle-associated pathways and elevated peptidase activities. Identification of these altered pathways exposes possible drug targets and, therefore, can help guide treatment decisions. Differential DNA methylation and miRNA signatures were also observed, which could explain the difference in transcriptomic and proteomic profiles of each subtype.
Lastly, the researchers selected a reduced set of 10 mRNA biomarkers and were able to predict drug responses of cell lines, that were not used in training of the model, with considerable accuracy. The reduced set of biomarkers increases this method’s utility in a clinical setting.
Sinkala et al.’s study is a great example of how informative molecular and genomic data can be for improving diagnostic, prognostic and treatment acumen in cancer. The advent of large-scale cell line databases has prompted a shift in the cancer treatment paradigm, away from a “one-size-fits-all” approach to personalised treatment, based on a patient’s unique biological and environmental context. This shift will hopefully see cancer falling from the second leading cause of death to a disease where poor outcomes are a thing of the past.
- Testing.com – What are genetic tests for targeted cancer therapy? [Internet]. [updated 2021 May 13; cited 2022 September 15]. Available from: https://www.testing.com/tests/genetic-tests-targeted-cancer-therapy/.
- Our World in Data – Causes of Death [Internet]. Hannah Ritchie and Max Roser. [updated 2019 December; cited 2022 September 15]. Available from: https://ourworldindata.org/causes-of-death.
- National Cancer Institute – Cancer Statistics [Internet]. [updated 2020 September 25; cited 2022 September 15]. Available from: https://www.cancer.gov/about-cancer/understanding/statistics#:~:text=Cancer%20is%20among%20the%20leading,related%20deaths%20to%2016.4%20million.
- Sinkala M, Mulder N, Martin D. Machine learning and network analyses reveal disease subtypes of pancreatic cancer and their molecular characteristics. Scientific reports. 2020 Jan 27;10(1):1-4.