Meno:Marek
Priezvisko:Kajan
Názov:Detection of ongoing cancer using DNA fragment characteristics
Vedúci:Mgr. Jaroslav Budią, PhD.
Rok:2025
Kµúčové slová:liquid biopsy, cancer, cfDNA, substitutions, population genomics
Abstrakt:Liquid biopsy is an emerging non-invasive diagnostic technique that detects tumor-derived material in blood, such as circulating tumor DNA (ctDNA). In this thesis, we investigate whether specific characteristics of cell-free DNA (cfDNA) fragments, particularly single nucleotide substitutions and their population rarity, can serve as reliable biomarkers for detecting ongoing cancer. Using two next-generation sequencing datasets (PreveLynch and GenoScan), we extracted substitution-related metrics from mapped cfDNA reads and analyzed their statistical distributions in healthy and oncology samples. We found that while population frequency of substitutions alone was not sufficiently discriminative, the overall distribution of substitution types showed statistically significant differences between the groups (p < 0.05 for most substitutions after Bonferroni correction). We engineered a set of 39 features and trained multiple machine learning classifiers. Classical models such as SVM and XGBoost achieved modest performance (ROC AUC = 0.70 and 0.55, respectively), whereas neural networks significantly outperformed them. The best performing model on PreveLynch dataset, a cross-attention neural network, achieved ROC AUC scores of 0.95. The best performing model on GenoScan dataset, a feedforward neural network, achieved ROC AUC scores of 0.85. These results demonstrate that substitution-based cfDNA fragment metrics contain meaningful signals that can aid in distinguishing cancer patients from healthy individuals. Our findings support the potential integration of such features into broader liquid biopsy frameworks for early cancer detection.

Súbory diplomovej práce:
Autor nedal súhlas so zverejnením svojej diplomovej práce.

Súbory prezentácie na obhajobe:

Upravi»