Implementation of machine learning and other simulation protocols for the representation of G-Quadruplex stabilizer

Abstract

Human chromosomal telomeres are capable of forming G-Quadruplex structures, which can inhibit the activity of telomerase, an enzyme commonly found in cancerous cells and largely responsible for their immortality. The inhibition may therefore lead to cell apoptosis in these cancer cells. As such, there have been growing interest in the research for G-Quadruplex stabilizers. This thesis focusses on the exploration and development of new representations for the structure-function relationship of G-Quadruplex-ligand biomolecular system using machine learning (ML) techniques. Two main models, namely Element Specific Persistent Homology (ESPH) and Rigidity Index-Score (RI-Score), were adapted due to their successes in representing other biomolecular systems. It was discovered that both methodologies similarly presented strong average positive correlation for the representation of the studied system. (Pearson Correlation: 0.6770-0.7871, RMSE: 0.4621-0.5811) In addition, most of the models developed performed admirably when compared to the well-established Quantitative Structure-Activity Relationship (QSAR) method. In particular, Ligand-based model using ESPH firmly outperformed all other models, new and existing alike. (Pearson Correlation: 0.7614-0.7871, RMSE: 0.4621-0.4866)