Symmetry Adapted Machine Learning Molecular Descriptors and Active Learning Approaches for Switchable Superlattices

This thesis presents advancements in computational methodologies to enhance molecular simulations through the development of active learning techniques, machine-learning methods, and robust open-source software engineering with a focus on ML-integration. The research is organized around four primary contributions: (1) computational modeling of self-assembling peptide--$\pi$ conjugated systems to guide the design of supramolecular nanomaterials, (2) the development and implementation of Permutationally Invariant Networks for Enhanced Sampling (PINES), (3) the high-throughput virtual screening of DNA-functionalized nanoparticles (DFPs) using an active learning approach, and (4) the structured software development of the PINES and ALPineFOREst frameworks.