Research Projects

Nucleic acids-DNA and RNA are dynamic biopolymers. Their three-dimensional (3D) structures are central to a wide array of essential cellular functions. These structures arise from a complex energy landscape shaped not only by canonical base-mediated interactions but also by electrostatic forces involving ion-mediated interactions, which are critical for stabilizing tertiary folds. These interactions minimize phosphate-phosphate repulsion in the nucleic acid backbone and enable long-range compaction. However, accurately predicting nucleic acid folding free energy landscapes and structures beyond the native basin remains a grand challenge in molecular biophysics and biophysical chemistry due to two major limitations: (i) the high computational cost of sampling long-timescale conformational and ionic interaction-associated dynamics, and (ii) the limited physical realism in current deep learning-based structure prediction models, which largely overlook the role of metal ions. This project proposes a novel hybrid framework- StEM-ɣFold that unites physics-based coarse-grained simulations with AI-guided RNA/DNA structure prediction. The central aim is to enable accurate prediction of folding pathways, ion-binding sites, and non-native metastable structures by capturing the influence of dynamic ions (e.g., Mg²⁺), which are essential for functional tertiary structures but often unresolved in experiments or ignored in AI models. StEM-ɣFold will be a transferable framework for nucleic acid structure exploration. The proposed work is organized under four major objectives: 1. Develop a Structure-based Electrostatic Model (StEM) of Nucleic Acid Fold that incorporates implicit monovalent ion condensation (using Dynamic Counter-ion Condensation theory) and explicit divalent ions (e.g., Mg²⁺). The model will use a hybrid Hamiltonian combining electrostatics, sequence-dependent energetics, and solvent effects. 2. Integrate Dynamic Ion-Exchange Phenomena by exploring the free energy profile of ions transitioning between outer/solvent-separated and inner-sphere/direct binding using umbrella sampling and atomistic simulations. This dynamic ion-binding potential will be embedded into StEM to predict folding under physiological ionic conditions and identify likely ion-binding sites. 3. Explore the Nucleic Acid Folding Landscape to predict functional but non-native conformers and intermediates. Folding trajectories will be analyzed using Time-lagged Independent Component Analysis (TICA), Markov State Models (MSMs), and umbrella sampling will be performed to navigate folding pathways and rate-limiting steps. 4. Develop StEM-Augmented Deep Learning Tools (StEM-ɣFold) by integrating ion-mediated constraints derived from StEM simulations and experimental data. Binary contact matrices of ion-mediated phosphate-phosphate interactions will guide and re-rank RNAformer outputs, closing the loop between simulation and AI. This integrated StEM-ɣFold framework will be validated on benchmark RNAs, including 58mer ribosomal RNA, the SARS-CoV-2 pseudoknot, and DNA G-quadruplexes. It will also be extended to circular RNAs (circRNAs), which are of emerging interest due to their stability and roles in cancer. Significance: This project addresses a key gap in simulation and AI-based nucleic acid structure prediction by incorporating ion-mediated electrostatics, an essential yet underexplored aspect of RNA/DNA folding. By modeling dynamic ion interactions, it will enable the discovery of novel, functional conformations with therapeutic potential. The work will advance India’s capabilities in AI-guided nucleic acid-based therapeutic designing and aligns closely with ANRF’s mission to promote frontier science and translational research.