Presentation: Integration of genomic data in the design of learning models in precision oncology
Acute lymphoblastic leukemia (ALL) accounts for nearly one-third of all pediatric cancers in Canada each year. Although current treatments achieve a 5-year survival rate of 90%, we observe that more than 65% of survivors develop long-term adverse effects including dyslipidemia, hypertension and osteoporosis. These effects contribute to the deterioration of the health status of these patients and can even constitute a danger for their lives. Our project consists in setting up a mathematical model allowing the prediction of these adverse effects, thus making possible a more adapted and personalized follow-up of ALL survivors. Several articles in the literature have highlighted the association between certain genes and the development of late adverse events. The use of genomic data could improve the quality of the predictions made by our model. The model will consist of a graphical neural network (GNN), an emerging architecture in the field of machine learning. It will include genomic data and other biomarkers from PETALE, a study of nearly 250 ALL survivors whose main objective was to identify and characterize the factors predisposing a survivor to develop health problems. The presentation aims at presenting the interest of GNN for the integration of genomic data in predictive models.