DEVELOPMENT OF VIRTUAL SCREENING METHODOLOGIES FOR IDENTIFYING RNA-BASED THERAPIES

This thesis focuses on the development of computational methodologies for predicting protein-RNA binding affinities and designing RNA sequences with specific functional outcomes. The research introduces PANTHER (Protein-Affinity for Nucleic Target-binding, Hybridization, and Energy Regression), a novel machine learning-based scoring function that predicts protein-RNA binding affinities through a local-to-global approach. The methodology integrates molecular dynamics simulations and machine learning models to estimate local interaction energies between amino acids and nucleotides, which are then aggregated to predict global binding affinities. A dedicated web service has also been developed to provide the public access of PANTHER score (https://nova.disfarm.unimi.it/panther/). In addition, the thesis extends the RISoTTo (RIbonucleic acid Sequence design from TerTiary structure) framework to handle conformationally flexible RNA molecules, enabling context-aware RNA sequence design. The integrated computational framework presented in this work aims to advance the understanding and practical prediction of protein-RNA interactions, supporting the development of RNA-targeted therapeutics and enhancing the field of computational structural biology.