/* Das ist der Code, damit das Akkordeon geschlossen angezeigt wird. */ /* Das ist der Code, um offene Akkordeons wieder schließen zu können */
Yifan Shi

Yifan Shi

Werner-Heisenberg-Gymnasium

 

Titel der Forschungsarbeit: DNA-Methylation Based Machine Learning Predictive Models for BRCAness detection in Osteosarcoma

School: TUM School of Medicine and Health

Department: Medizin

Lehrstuhl: Forschergruppe für Molekulargenetische Veränderungen beim Osteosarkom

Betreuung: Dr. Maxim Barenboim

Abstract der Forschungsarbeit

The effective administration of Poly-ADP-ribose inhibitors (PARPi), a promising therapeutic approach for cancer treatment, depends on the presence of homologous repair deficiency (HRD) in tumour cells. BRCAness describes the phenotype where somatic mutations of BRCA1/2 or other genes involved in the HR pathway are present, resulting in HRD. Osteosarcoma (OS), a prevalent paediatric bone cancer, exhibits a tendency to develop BRCAness over time, making it a potential target for PARPi therapy.

This study aims to improve the predictive accuracy of using DNA methylation data to detect BRCAness in OS. Various machine learning algorithms were applied to this problem to find the one that predicts BRCAness in OS most accurately compared to the original Random Forest (RF) classifier. To assess the performance of the algorithms, the Waikato Environment for Knowledge Analysis (WEKA) was utilised for initial comparison. Subsequently, the algorithms were implemented in R and their performance evaluated using the area under the receiver operating characteristic (ROC) curve. The closer the area under this curve (AUC) to 1, the higher the performance of the algorithm. It was revealed that multilayer perceptron (MLP) outperformed RF, achieving an AUC of 0.96 compared to RF’s 0.87. Logistic regression (LR) and Naïve Bayes (NB) also demonstrated superior performance, with AUCs of 0.93 and 0.9, respectively. Furthermore, the positive effect of increasing the training set size on machine learning model performance was confirmed and BRCAness was detected in 13 new OS samples. These findings highlight the potential of MLP, LR, and NB as more accurate predictive models of BRCAness in OS.