Proteomics analysis of septic and cardiogenic shock

Treballs finals del Màster de Fonaments de Ciència de Dades, Facultat de matemàtiques, Universitat de Barcelona, Any: 2018, Tutor: Vicent Ribas Ripoll[en] Introduction: Shock is a life-threatening condition affecting about one third of patients in the ICU. The most common types of Shock are Septic and Cardiogenic, affecting 62% and 16% of Shock patients, respectively. A rapid and specialized treatment focused on the type of Shock is crucial for reducing its high mortality rate. Unfortunately, current therapies strive to reduce the medical signs present by the patients rather than target the cause of Shock. A better understanding of the mechanisms and pathophysiology of Shock is mandatory for improving its diagnosis. Omic data and Machine Learning techniques make the perfect combination to tackle this challenge. Methodology: In this thesis, a two-step Machine Learning model has been proposed for analysing proteomic data. The model consists of a Feature Selection method, aimed at selecting relevant proteins, followed by a Classification method, whose purpose is to predict the type of Shock. A robust procedure has been designed for selecting the best model, i.e., stable, interpretable and accurate. Since there is no consensus on the best stability measure, an analysis of different metrics has been performed to decide which metric is more suitable for our problem. Conclusions: Promising results have been obtained using the proteomic data collected in the European research project ShockOmics from Septic and Cardiogenic Shock patients. The best model, a combination of ReliefF and Random Forest, is capable of perfectly discriminate between these two types of Shock. On top of that, the proposed model selected meaningful proteins which have been extensively studied in the literature for its relation with Septic Shock