Molecular Geometry Generation Processes Through Hybrid Quantum-Classical Generative Adversarial Networks and Python-Based Self-Consistent Field Molecular Calculations

Current compound synthesis and reaction pathway determination are implemented with computational chemistry methods, but methodological improvements are needed to increase precision, accuracy, and efficiency. We are implementing a molecular geometry generation process using Hybrid Quantum-Classical Graph Generative Adversarial Networks (QNetGAN), STO-nG basis set, Hartree-Fock (HF) and Post-Hartree-Fock (PHF) approximations of Schrödinger's equation for molecular energy calculations using PySCF. QNetGAN generates molecular graphs, ensuring atom connectedness, druglikeness, and Octet Rule satisfaction, at a fast runtime and great efficiency. Although a work in progress, QNetGAN has demonstrated its capabilities with an impressive training time of 10.164 minutes and 47% success rate, generating 141/300 structurally valid molecular geometries that fulfill Lipinski's Rule of Five. Meanwhile, we have developed a Python-based molecular energy and properties program with proven accuracy. By testing pre-existing organic molecules using HF (and later, PHF) and comparing our experimental results with accepted values from the CCCBDB government database, we are verifying the reliability of our program and preparing to determine the stability and feasibility of future QNetGAN-generated molecules. Currently, the QNetGAN and the HF/PHF molecular geometry and molecular energy calculations are separate; future work involves improving the accuracy of QNetGAN before integrating these components together and the current classical GAN/Density Functional Theory (DFT) systems. The ultimate goal is to increase efficiency and lower costs in R&D processes, accelerating the development of new medicines and treatments for currently incurable diseases.