My Work 

Abstract: Molecular property prediction is a central task in quantum chemistry and drug discovery, traditionally requiring expensive ab initio calculations. Recent advances in graph neural networks (GNNs) have enabled data-driven approaches to predict quantum mechanical properties of molecules with high accuracy. In this paper, we leverage the QM9 dataset of small organic molecules (with computed geometric, energetic, electronic, and thermodynamic properties) to evaluate GNN models against classical machine learning methods. We find that GNN-based models significantly outperform linear regression, support vector regression, and random forest baselines in predicting molecular properties such as dipole moment, polarizability, and frontier orbital energies. Notably, state-of-the-art GNNs achieve mean absolute errors approaching chemical accuracy (within 1 kcal/mol) for energies, offering insights into molecular structure–property relationships. Our results demonstrate that incorporating molecular graph structure via GNNs, combined with quantum mechanical data, yields superior predictive performance. These findings highlight the potential of GNNs to accelerate virtual screening and discovery by providing fast, accurate predictions of quantum-derived molecular properties.

Abstract: We report first-principles phonon dynamics studies in isotropically strained mono-layer silicene films of various topologies by density functional theory and density functional perturbation theory methods. The strain introduced in the sheets are of both tensile and compressive nature and its effect on the thermal transport properties are analysed using the Boltzmann Transport Equation (BTE) reported. We performed electronic and vibrational analysis of each silicene topology and concluded that hhk-silicene is a potential candidate for next generation electro-thermal devices due its molecular stability, bulk-Si like bandgap and tolerable thermal performance that can be engineered by strain. The novel hhk-silicene topology offers excellent thermal and electronic capabilities that are sensitive to strain along with the band gap following a near-linear monotonically decreasing trend making it suitable for strain-tronics applications.