Laezza, Luigi (2024) Somatic Mutations Analysis in Different Genetic Ancestries across Cancers. [Tesi di dottorato]

Documento PDF CQB_Thesis_37cycle_final.pdf Visibile a [TBR] Amministratori dell'archivio Download (12MB) | Richiedi una copia

Abstract

In precision oncology distinguishing between germline and somatic genetic mutations has become an integral part of diagnosis and treatments. Accounting for patients genetic ancestry further improves outcomes by considering ancestral influence on mutation prevalence and drug response. This thesis investigates the integration of advanced computational methods with genetic data to enhance the detection of the DNA mutations underlying cancer and contribute to provides analytic support to genetic ancestry influence on cancer disparities. First, a robust machine learning classifier was developed and implemented to accurately discriminate between germline and somatic DNA variants in tumor-only whole exome sequencing data. Second, a genetic ancestry detection method was developed to properly categorize high admixed ancestry samples. Third, appropriate statistical methods, such as multivariate logistic regression models and ${\chi}^2$ tests, were customized to recognize the effect of genetic ancestry on somatic mutations on both the pan-cancer and cancer-specific levels. Our optimized variants classifier, trained on 200 patients across 13 diverse tumor types from The Cancer Genome Atlas, consistently demonstrated high performance in cross-validation, achieving a ROC-AUC of 96% ± 2 and a PR-AUC of 92% ± 5. The model also performed well on an external validation cohort from the University of Miami, enabling reliable analyses on the full dataset. Ancestry detection of this cohort revealed significant representation of Admixed American, European, and African ancestries. The ancestry-based somatic mutations analysis revealed multiple associations, including a higher frequency of KRAS mutations in European lung adenocarcinoma patients and PIK3R1 mutations in Admixed American uterine corpus endometrial carcinoma patients.

Downloads

Actions (login required)

Modifica documento