Stratification of patient subgroups using high-dimensional and time-series observations
Neyton, Lucile Patrick Antoine
Precision medicine and patient stratification are expanding as a result of innovations in high-throughput technologies applied to clinical medicine. Stratification can explain differences in disease trajectories and outcomes in heterogeneous cohorts. Thus, approaches employed for patient treatment can be tailored by taking into account individual variabilities and specificities. This thesis focuses on clustering approaches and how they can be applied to both single time points and time-series high-dimensional data for the identification of disease subtypes defined by distinct mechanisms, also called endotypes, in complex and/or heterogeneous diseases. Multiple carefully selected clustering strategies were compared to highlight which would produce the most relevant stratification in terms of mathematical robustness and biological meaning, both of which quantified using standardised methods. More specifically, this strategy was applied to time-series multi-omics data from a cohort of patients with acute pancreatitis, an inflammatory disease of the pancreas. Using this high-dimensional multi-omics data as well as routine lab and clinical measurements, the cohort was stratified into four subgroups. Findings from the analysis of acute pancreatitis data showed that two of the four subgroups could be detected in another syndrome, acute respiratory distress syndrome, suggesting that inflammatory signatures are comparable between diseases. With the aim of applying these principles to other diseases and using preliminary results from other studies suggesting that relevant subgroups might be highlighted, data from inflammatory bowel disease and Parkinson's disease cohorts was analysed. Results from our analyses confirmed that disease knowledge could be gained using this approach. Work from this thesis provides novel approaches for the application and evaluation of stratification methods. Furthermore, results may constitute a basis for the development of tailored treatment approaches for acute pancreatitis, acute respiratory distress syndrome, inflammatory bowel disease and Parkinson’s disease. Also, the observation of commonalities between distinct inflammatory diseases will broaden the perspectives when analysing disease data and more specifically, in biomarker discovery and drug development processes.