Doctoral Exit Seminar: Machine learning models for molecular based functional organic materials
Organic semiconductors (OSC) are of interest for a wide range of flexible optoelectronics applications, including transistors, solar cells, and sensors, to name a few. Despite their promise, the design and optimization of OSC pose significant challenges due to the complexity of the structures of the molecular building blocks, varied packing configurations of these building blocks in the solid state, which impacts the optical and electronic response, and sensitivity of the solid-state packing to material processing conditions. Accurately predicting the solid-state properties of OSC traditionally requires high-level quantum mechanical methods. These methods, however, can be computationally demanding, particularly for large molecules or when there is interest in extensive material screenings. Overcoming this computational bottleneck is essential to enabling the efficient design of OSC, which would reduce the experimental trial-and-error approach used in material discovery. Moreso, the holy grail of computational study is to be able to accurately and efficiently predict the molecular packing configurations and associated properties of OSC. This dissertation aims to address some of these challenges by developing computational approaches that leverage machine learning (ML) models to accelerate the study of OSC. ML promises to facilitate faster material screening and optimization by offering an alternative to direct quantum mechanical calculations. Specifically, this dissertation describes the development of ML models for intermolecular interactions, including noncovalent interactions (NCI) and electronic couplings (EC). Conventional quantum mechanical methods used to investigate OSC are introduced, and ML approaches are reviewed. The dissertation then discusses the generation of large, high-quality datasets for NCI from symmetry-adapted perturbation theory (SAPT), and the development of ML models to efficiently predict NCI. An active learning approach for the high-throughput derivation of optimal training sets for NCI predictions is then developed, and the training set is used to train new ML models. Finally, ML models to predict EC from three-dimensional (3D) molecular dimer geometries are implemented for the rapid, on-the-fly prediction of ECs across thermally sampled conformations obtained through molecular dynamics (MD) simulations to enable rapid materials characterization during simulation. Ultimately, this dissertation presents a framework that integrates ML with quantum mechanical insights, offering a scalable solution to accelerate OSC discovery and optimization.
KEYWORDS: Organic Semiconductors (OSC), Density Functional Theory (DFT), Symmetry-Adapted Perturbation Theory (SAPT), Noncovalent Interactions (NCI), Electronic Couplings (EC), Machine Learning (ML).
Cerebrovasculature refers to the network of blood vessels in the brain, and its coupling with neurons plays a critical role in regulating ion exchange, molecule transport, nutrient and oxygen delivery, and waste removal in the brain. Abnormalities in cerebrovasculature and disruptions of the blood supply are associated with a variety of cerebrovascular and neurodegenerative disorders. Nanocarriers, a nano-sized drug delivery system synthesized from various materials, have been designed to encapsulate therapeutic agents and overcome delivery challenges in crossing the blood-brain barrier (BBB) to achieve targeted and enhanced therapy for these diseases. Unraveling the transport of drugs and nanocarriers in the cerebrovasculature is important for pharmacokinetic and hemodynamic studies but is challenging due to difficulties in detecting these particles within the circulatory system of a live animal. In this dissertation, we developed a technique to achieve real-time
The health of our communities depends on the effective treatment of both solid and liquid waste to eradicate hazardous pollutants before they can interact with living organisms or contaminate the environment. Daily, society generates solid waste (commonly destined for landfills) and liquid waste, (commonly discharged into wastewater systems) and without proper treatment, these wastes can release hazardous primary secondary pollutants. Industries producing wastewater with high pollutant concentrations, especially those utilizing lignin-based biomass, face complex challenges because each facility may require a tailored treatment approach. In response, this work investigates the use of ozonolysis to transform lignin monomers into smaller, less hazardous components that can be more efficiently managed by public wastewater systems. Furthermore, while conventional wastewater treatment systems are effective for common water quality issues, they can inadvertently allow complex compounds, such pollutants from hospital effluent, to pass through. Under simulated treatment conditions incorporating sunlight and chlorination, a pollutant released from medical facilities is degraded, but this process may also lead to the formation of carcinogenic disinfection by-products (DBPs) that pose direct toxicological risks to nearby communities. The implications extend to solid waste management as well. Chemical phenomena, such as those occurring in poorly understood elevated temperature landfills (ETLFs), can compromise treatment methods and increase community exposure to harmful pollutants. By monitoring hazardous components, such as volatile organic compounds (VOCs), over time, this work aims to elucidate the chemical reactions occurring both during treatment and in the environment thereafter. Ultimately, this research underscores the need for fundamental, innovative approaches to pollution transformation. Bridging the gap between existing practices for solid and liquid waste treatment will be critical to safeguarding environmental and public health.
