Application of Machine Learning Techniques to Acute Myeloid Leukemia
This thesis is inspired by the position paper ”Predictive, personalized, preventive, par- ticipatory (P4) cancer medicine” . The basic concept of P4 medicine was “The right patient with the right drug at the right dose at the right time.”. In other words, the goal is to tailor medical treatment to the individual characteristics, needs, and preferences of a patient during all stages of care, including prevention, diagnosis, treatment, and follow-up . In this thesis, we used Acute Myeloid Leukemia (AML) as our case study because if untreated, AML progresses rapidly and is typically fatal within weeks or months, and also because genomic data were available. It has also been shown that AML is associated with gene mutations , and hence, genomic approaches have the potential to contribute to this heterogeneous cancer. We applied machine learning algorithms to build predictive models using biomedical data profiling AML patients. Specifically, in chapter 1 we introduced the problem and related works of our projects and in chapter 2 we introduced background knowledge of all the algorithms and technologies being used. In chapter 3, we report the identification of 24-gene signature predictive of the relapse of low-risk AML patients. These 24 genes could be used to distinguish a future patient will be relapse or non-relapse. Our findings in chapter 3 were derived by mining gene expression data of AML patients generated from microarray technology and next generation sequencing technology. We would like to note the limitations of this ”personalized medicine” approach: further clinical evidence and trials would be needed to elucidate the underlying biological mechanisms. In chapter 4, we applied correlation analyses to high-throughput drug sensitivity data to identify gene mutations that could be potential candidates to explain patients’ responses to AML drugs. In chapter 5, we concluded our projects and given an overview of possible future works. In this thesis, we focused on AML as our case study. However, our methods could be applicable to other diseases for which data are available.