Project description
Harnessing statistical analyses to improve multiple myeloma outcomes
Multiple myeloma is a type of bone marrow cancer affecting white blood cells called plasma cells. These cells help the body fight infections by making antibodies that recognise and attack invaders. In multiple myeloma, cancer cells that produce abnormal proteins accumulate in the bone marrow, destroying bone and displacing healthy blood cells. The cancer can affect multiple areas of the body, including the spine, skull, pelvis and ribs, hence the term 'multiple myeloma'. The relapse rate is 100 % – it is not curable but can go into remission. However, treatments are available. The EU-funded LIMORD project is developing a statistical tool that will be incorporated into a software package for clinical use, enabling more accurate patient classification, earlier relapse detection and better prognosis estimation. The tool will support personalised medicine and improve patient outcomes.
Objective
The goal of my project is to propose a novel statistical tool allowing patient classification, earlier relapse detection and better prognosis estimation in order to move forward into personalized medicine in Multiple Myeloma. To this aim, I will develop new statistical models and computational schemes to incorporate large follow-up omics datasets in a decision framework. As a statistician coming from theoretical mathematics, this project will provide me a unique opportunity to acquire new knowledge in biology and new supervision skills in order to translate theoretical mathematical results into real added value in the way we treat patients.
The first challenge I will address is the development of statistical methods based on Variational Auto-Encoders to integrate multiple omics data-type at multiple time-points. My model will have to be flexible enough to allow for missing data (for instance a full omic dataset missing at a given time point due to experiment failure) and to accommodate for data acquired in an online manner. The second challenge I will address is the development of quality metrics and analysis methods for direct RNA sequencing data from patient samples. The third challenge I will address concerns the numerical inference difficulties of Partially Observable Markov Decision Processes when the dimension of the data increases. Approximation strategies will be investigated to make use of the high-dimensional, heterogeneous biological data in a relapse detection framework. Finally, I will develop a software package incorporating our results intended to help clinicians take the optimal decision when treating their patients.
An important aspect of my project is to integrate it both to a biological laboratory in Australia and a mathematical group in France, together with a collaboration with clinicians in a French hospital, hence I will carry out the entire process of designing the statistical tool and its software package for a concrete use in the clinic.
Fields of science
- natural sciencescomputer and information sciencessoftware
- natural sciencescomputer and information sciencesartificial intelligencemachine learningreinforcement learning
- medical and health scienceshealth sciencespersonalized medicine
- natural sciencesmathematicsapplied mathematicsstatistics and probability
- natural sciencesbiological sciencesgeneticsRNA
Programme(s)
Funding Scheme
MSCA-IF - Marie Skłodowska-Curie Individual Fellowships (IF)Coordinator
75794 Paris
France