Discovery of RNA-Targeted Small Molecules by Quantitative Structure-Activity Relationship (QSAR) Study and Machine Learning

Zhengguo Cai, Ph.D. Candidate Amanda Hargrove Ph.D., Advisor Abstract: The diversity of RNA structural elements and their documented role in human diseases make RNA an attractive therapeutic target. However, the distinct chemical properties of RNA molecules, such as the densely charged backbone, the structural flexibility and the limited chemical diversity of building units, lead to a significantly different ligand space when compared to protein peers, as well as difficulties of transferring protein-based methodology to discover RNA-targeted small molecules. In the past few years, I have developed a general workflow of quantitative structure-activity relationship study (QSAR) to understand the molecular factors that drive RNA recognition by small molecules. Introduction of Kennard-Stone sampling, least absolute shrinkage and selection operator (lasso)-based feature shrinkage and multiple linear regression to the experimental data afforded predictive models as well as explicit interpretation of the contributing parameters. The developed QSAR models have provided predictions of small molecule candidates and informed the medicinal chemist for the synthesis of potential ligands targeting different RNA structures. In addition, it was the first time that the QSAR model and deep learning generative model (e.g., variational auto-encoder) was combined to automatically generate novel ligands with desired RNA binding properties. These computational methods could greatly enrich the current arsenal for the development of RNA-targeted small molecules and explore the untapped chemical space efficiently.