DESIGN AND DEVELOPMENT OF RNA-BASED DRUGS USING MOLECULAR DYNAMICS SIMULATIONS AND ARTIFICIAL INTELLIGENCE METHODS

RNA-based therapeutics represent a powerful class of drugs offering precise targeting for genetic disorders, cancer, and infectious diseases. Several RNA-based modalities have been developed, including messenger RNAs (mRNAs), small interfering RNAs (siRNAs), microRNAs (miRNAs), antisense oligonucleotides (ASOs), circular RNAs (circRNAs), and aptamers. Despite their significant potential, these therapeutics face important limitations, including susceptibility to RNase degradation, high net negative charge, and rapid renal clearance. To address these challenges, advanced chemical modifications, such as Xeno Nucleic Acids (XNAs) and 2-OH modifications, along with specialized delivery systems, including viral vectors and lipid nanoparticles (LNPs), have been introduced. The design of these advanced ligands increasingly relies on computational approaches, including structure-based drug design (SBDD), molecular dynamics (MD) simulations, and machine learning (ML). This thesis presents a comprehensive in silico framework for the design and optimization of nucleic acid therapeutics targeting three distinct systems. First, autophagy inhibition was pursued by targeting the LC3B RNA-binding motif using novel Peptide Nucleic Acids (PNAs) to disrupt cancer cell survival mechanisms. Second, XNA sequences were optimized to target the DNA-binding domain of the oncogenic protein HMGB1. Third, aptamers were designed to bind the CD4-binding site of the HIV-1 envelope glycoprotein GP120, thereby blocking viral entry. All projects followed a multi-step computational and experimental pipeline. Protein structures retrieved from the Protein Data Bank (PDB) were prepared and used for docking studies to identify ligand binding sites. Subsequently, MD simulations were performed to evaluate the stability of ligand–target complexes, and binding free energies were estimated using MM/GBSA methods. Additionally, an Active Learning (AL) model was developed to analyze multi-dimensional (1D–4D) molecular descriptors for ligands targeting LC3B. This workflow was adapted and optimized for each target system. The results demonstrate that targeting the LC3B RNA-binding motif can disrupt LC3B–mRNA interactions, preventing lysosome-mediated degradation of tumor suppressors such as PRMT1 and limiting cancer cell survival. Microscale thermophoresis (MST) experiments on the PNA ligand AATAAA revealed a dissociation constant in the low nanomolar range. For HMGB1, sequence optimization of the initial ATAG motif led to a twofold improvement in predicted binding affinity. The lead Tyr2 analog of GAGG-opt exhibited a binding free energy of –149.3 ± 6.3 kcal/mol and a Kd of 26.9 ± 1.36 µM against the HMGB1 A-box domain. In the HIV-1 system, the designed aptamers targeting the GP120 CD4-binding site highlighted the critical role of sequence length and composition in achieving high affinity. Furthermore, glycan interactions were essential for aptamer stability, and backbone optimization was strongly dependent on both sequence and folding. In conclusion, this work demonstrates that the integration of computational chemistry, machine learning, and experimental validation provides an effective strategy for the rational design of nucleic acid therapeutics targeting complex biological systems.