Predicting clinical outcomes employing support vector machine and assessment of mouse model validity
This thesis employs computational and experimental methods to identify predictive and functional biomarkers in three contexts. By integrating transcriptomic datasets with machine learning (ML) techniques, using linear support vector machine (SVM), and systems biology approaches, it evaluates data-driven strategies strategies in personalized medicine and assesses murine model validity for microbiome-related functions. In the first study, peripheral blood mononuclear cells (PBMCs) were sampled from children with short stature before and four days after recombinant human growth hormone (r-hGH) treatment. Linear SVM models based on just five genes achieved a high median balanced accuracy in predicting long-term growth outcomes 4 years later. Furthermore, the predictions exhibited a prognostic value beyond 4 years. The second study re-contacted a cohort of patients with atopic dermatitis 10 years after initial enrollment to collect long-term clinical outcomes. Combining this clinical information with transcriptomic profiling of lesional skin, non-lesional skin and blood, linear SVMs classifiers categorize patients by parameter oSCORAD, EASI, itch (3d-NRS) and flares. Although models showed high prediction accuracy, skin-derived signatures did not transfer effectively to blood, and expression quantitative trait locus (eQTL) analysis failed to link genetic variants with the identified predictors. The third study exposed mice totobacco smoke, then analyzed their gut microbiota composition and host transcriptomes. A comparative analysis of transcriptomic data from mice and humans was conducted, with the integration of microbiota data from male and female mice exposed to smoke and air. The results demonstrated shared microbial functional pathways alongside model-specific differences. This finding validate murine models for studying many human microbiome-host interactions, keeping in mind the limitations detected in specific pathways.
Rights
Use and reproduction:
Please note that individual components of the publication may be subject to other licensing or copyright conditions.