De Novo Structure Prediction of LncRNA MALAT1

Abstract

Functional, noncoding RNAs (ncRNAs) have come to light in recent years for their key role in biological regulatory processes and association with notoriously difficult to treat diseases such as cancer. The long ncRNA MALAT1, a potential cancer biomarker and target for cancer treatment, is studied using novel computational methods. However, despite growing interest in RNA as a therapeutic target, determining their structure is difficult. Current computational techniques struggle to understand RNA structure due to highly complex RNA dynamics occurring over long timescales and limitations in addressing non-canonical base pairing. Moreover, RNAs are flexible by nature and adopt a broad ensemble of conformation rather than a single native state. Through the development of Thermodynamic Maps, which combines statistical mechanics and molecular simulations with score-based generative models, a sequence-to-structural ensemble pipeline is created to perform de novo structure prediction of MALAT1 and yield an enriched structural ensemble at hard-to-sample physiological temperatures. Whilst tackling challenges such as the large size of MALAT1 which greatly affects sampling and incorporating extensive exploration of Langevin dynamics, essential to score matching, the key mechanistic functions of the RNA are able to be studied with the discovery of its ensemble. With support from the National Cancer Institute, work continues to aim to find small molecules that emerge as potential ligands to MALAT1, pushing towards drug discovery and optimization.