A Predictive Model for Monolayer-Selective Metal-Mediated MoS2 Exfoliation Incorporating Electrostatics
A. Corletto, M. Fronzi, A. K. Joannidis, P. C. Sherrell, M. J. Ford, D. A. Winkler, J. G. Shapter, J. Bullock, A. V. Ellis, A Predictive Model for Monolayer-Selective Metal-Mediated MoS2 Exfoliation Incorporating Electrostatics. ADVANCED MATERIALS INTERFACES 2024, 11, 2300686.
https://doi.org/10.1002/admi.202300686
2022-2023
-
Deep Neural Network Enhanced Stochastic Optimization of Multistep Enzyme Catalyzed Industrial Codeine Bioproduction
Final Project - System Optimisation and Machine Learning ELEN90088
The multistep enzyme catalyzed bioconversion of codeine from thebaine presents a more ecologically friendly route for codeine production, but industrial adoption hinges upon achieving economic viability, necessitating optimization of the process. To achieve this, maximum codeine production for the process was framed as an optimization problem, setting reaction parameters including enzyme introduction time, temperature and pH, as decision variables, and simulated with a python kinetics model.
Enzymes, as complex biological molecules, are prone to deviation from their expected performance, and hence the role of enzymes in this bioconversion embeds uncertainty into the optimization problem. This uncertainty was mitigated by development of a 2-step stochastic optimization method, whereby initial reaction parameters were optimized according to expected kcat values - representing the expected enzyme efficiency of the system. After an initial, incomplete period of reaction, a deep neural network model was utilized to predict the true kcat values from the incomplete reaction kinetics time-series data, allowing the incomplete reaction to be reoptimized according to this more accurate data. A 2.6% average increase in codeine production over the non-reoptimized process was observed for the stochastically reoptimized process.
Support vector machine models were compared to deep neural networks for several reaction kinetic data representations for detection of competitive contaminant inhibition of the steps of the reaction. Both methods performed similarly, with both performing best using PCA reduced Euclidean distance representations of the kinetics time series data (~95% detection accuracy).
2023
-
New Asymmetric [CuATSM] Analogues for Neurodegeneration Diagnosis and Therapy: Donor Atom Effects on Redox Potential and Reversibility
Research Project, Donnelly Laboratory, Bio21 Molecular Science and Biotechnology Institute, The University of Melbourne
[CuATSM] is a copper-based compound undergoing clinical trials for Parkinson's disease and MND, with potential for the diagnosis and treatment of these and Alzheimer's disease. Next generation variants, focused on two of its proposed mechanisms of action were developed, with the aim of improving understanding and/or efficacy of its activity.
2020-2021