Unravelling the interaction between somatic genetics and treatment response in leukaemia using machine learning

Abstract

Acute lymphoblastic leukaemia (ALL) is the most common type of cancer affecting children with a peak prevalence between the ages of 2 and 5 years old. Development of effective treatments and improvements in risk stratification has led to a cure rate >90% in children. However, there are significant long-term effects associated with treatment of ALL. As a result, current research efforts have focused increasingly on identifying patients eligible for treatment de-escalation. Recent studies suggest that modest de-escalation of treatment for low risk patients is safe, namely patients with low levels of MRD and good risk genetics (ETV6::RUNX1 and High Hyperdiploidy). The objectives of this project were to identify optimal treatment elements for patients with good risk genetics to ensure that these patients are given only the minimal dosages of drugs necessary to be cured. The survival rates of good risk genetics (ETV6::RUNX1 and high hyperdiploidy) patients across four UKALL trials were determined, and the impact of different treatment elements was assessed individually using traditional statistical techniques to identify optimal treatment elements for these patients. Individual drug dosages were calculated using the trial protocols and a clinically annotated dataset (n = 6716) was assembled from both this information and data from LRCG sources. Area under the curve (AUC) was used to produce a dose intensity score (DIS) which was utilised as a method for determining optimal drug dosages for patients. Machine learning methods were explored with classification decision tree models being produced as well as ensemble methods being employed to identify optimal treatment elements within the aforementioned trials and analyse treatment effect on survival. In conclusion, successful treatment pathways that optimise outcome and minimise toxicity exist within historic clinical trials. Furthermore, optimal doses of drugs given on current treatment protocols have been identified for good risk ALL patients.