(118i) High-Throughput Screening of Tribological Properties of Monolayer Films Using Molecular Dynamics and Machine Learning

Monolayer films are promising candidates as lubricants for nanoscale devices in which sliding surfaces are in contact. In such systems traditional lubricants have been shown to have limited effectiveness as they can undergo a phase transition (from liquid to solid) when confined at the nanoscale [1, 2]. Monolayer films consisting of chains chemisorbed to a surface have been shown to provide protection against wear and reduce frictional forces in the event of contact. Various tunable properties of the monolayer including terminal group chemistry, backbone chain composition as well as monolayer composition have an impact on the lubricating ability of these films [2]. However, the parameter design-space is vast and requires significant computational resources to study. To address this, we have conducted a high-throughput screening study using molecular dynamics (MD) simulations and machine learning (ML) to determine optimal monolayer film designs, focusing on terminal group chemistry and monolayer composition. Approximately 120,000 simulations have been performed, representing nearly 10,000 unique monolayer film combinations. The cheminformatics lib RDKit, which is used to determine the “molecular fingerprint” of each system [7], and random forest regressor machine learning algorithm was used to create predictive ML models from the simulation data [8]. The effectiveness of the ML models and the optimal amount of training data is assessed. Several promising systems to be further investigated were identified highlighting how ML models can be used in lieu of simulation data to direct screening studies [3]. Such early evaluations of thin film designs can speed up the discovery process by minimizing the number of compute-intensive simulations that need to be performed. The MD simulations and high-throughput screening workflow was set up using the MoSDeF framework, ensuring the simulations are TRUE (transferable, reproducible, usable by others and extensible) [9]. The Signac framework was used to create and manage the parameter space and control the simulation workflow [5].

References

1. R. Maboudian, W. R. Ashurst, and C. Carraro. Tribological Challenges in Micromechanical Systems. Tribology Letters 12, 95–100 (2002). https://doi.org/10.1023/A:1014044207344
2. P. T. Cummings, H. Dochert, C. R. Iacovella, and J. K. Singh. (2010), Phase transitions in nanoconfined fluids: The evidence from simulation and theory. AIChE J., 56: 842-848. https://doi.org/10.1002/aic.12226
3. Y. L. Pavlov. (2019) ‘Random forests’, Random Forests, pp. 1–122. doi: 10.1201/9780429469275-8.
4. “MoSDeF” [Online]. Available: https://mosdef.org.
5. “Signac Framework” [Online]. Available: https://signac.io.
6. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl. “GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers”. SoftwareX, Volumes 1–2, 2015, Pages 19-25. https://doi.org/10.1016/j.softx.2015.06.001.
7. “RDKit Library” [Online]. Available: https://github.com/rdkit/rdkit.
8. “Scikit-learn” [Online]. Available: https://github.com/scikit-learn/scikit-learn
9. M. W. Thompson, J. B. Gilmer, R. A. Matsumoto, C. D. Quach, P. Shamaprasad, A. H. Yang, C. R. Iacovella, C. McCabe and P. T. Cummings (2020) Towards molecular simulations that are transparent, reproducible, usable by others, and extensible (TRUE), Molecular Physics, 118:9-10, DOI: 10.1080/00268976.2020.1742938