Comparative study of statistical methods for finding biomarkers in longitudinal data

Solid organ transplantation is a common procedure for end-stage organ failure. After the transplantation, the rejection of the new organ is possible due to the patient's immune system trying to eliminate the foreign object. To prevent rejection, monthly painful and costly procedure is needed which involves taking a biopsy of the allograft. The purpose of our project is to find biomarkers based on blood samples, so the diagnosis/prognosis of the rejection can occur based on a simple blood or urine sample. Up to eight blood samples are taken from rejection and non-rejection patients and for each sample a microarray is created. The microarray data is longitudinal and contains 54,613 genes. The analysis consists of normalization, pre-filtering, filtering, testing the candidate biomarkers for diagnosis/prediction, pathway analysis, and biomarker validation. For our type of data, the bottleneck and the most understudied step is the filtering. We focused our research on finding possible filtering methods. We tested these methods against the questions our biologists wanted to get the answer for. We generated a data set, based on real data, to find the strengths and weaknesses of the filtering methods we proposed to use. We also tested which one of the filtering methods would provide the most precise answer to each group of questions by creating synthetic data sets with a number of biomarkers planted in them. Our conclusion is that a statistical method, or group of methods, would not be able to provide the perfect answer to all of our biological questions. That is why we created a table where we matched our questions to methods that, based on our experiments, give the best results. Also, we provided some advice on which methods perform better under specific conditions.Science, Faculty ofComputer Science, Department ofGraduat