| dc.description.abstract | Monolayer film coatings show promise as a means to lubricate mechanical devices with nanoscale separation. These monolayers have many optimizable parameters, such as its film composition and terminal groups, that can significantly improve their lubricating efficacy. However, the vast screening space presents a unique challenge, which required the development of an efficient approach to screen for promising candidates. This work utilized the Molecular Simulation Design Framework (MoSDeF) and Signac Framework to design a molecular dynamics high-throughput screening workflow. The workflow was employed to explore 9747 monolayer film configurations, with 116,964 simulations, and was able to determine a shortlist of configurations that offer advantageous lubricating effect. The in silico data used to train machine learning (ML) models used a random forest regressor algorithm. The models offered insight about the quantitative structure-property relationship (QSPR) between the terminal group and the final tribological effectiveness of the monolayer. This helped extrapolate from trends found in the surveyed parameter space, allowing for predictions of monolayer configurations unaccounted in this work. The predictive models were applied to screen over 193,131 unique film candidates generated from the ChEMBL small molecule library, offering a 5 order of magnitude speed up compared to MD simulations. With the available data, we also explore the hypothetical scenario where ML can be integrated into the high-throughput screening process to significantly improve its efficiency. This work shows that the combinatorial approach utilizing both MD and ML can significantly improve the high-throughput screening workflow. | |
