Benchmarking quantum against classical schemes for Machine Learning based analysis of protein simulations

The methodology applied is related to unsupervised Machine Learning including dimensionality reduction and clustering schemes that are being compared across different classical, quantum simulator, and quantum architectures. No specific algorithmic development is being performed in this work, rather a benchmark and comparison of existing schemes on different platforms. Our aim is to validate existing methodologies and assess the applicability of these across the platforms in the field of biomolecular simulations rather than prepare quantum registers. For this, we use existing platforms in order to assess the quality of the Machine Learning based analysis on simulation data and focus rather on performance and accuracy.

The practical demonstration is related with the testing of different platforms with respect to the application of unsupervised Machine Learning schemes. Specifically, focus is given on classical dynamics simulations of cyclic proteins in solution and the analysis of the simulated data on the different computing platforms. For this existing algorithms are being applied with the focus to test their accuracy, performance, and overall applicability on the different platforms. To this end, different simulated data, such as structural properties of the proteins are analysed with the aid of unsupervised Machine Learning in order to demonstrate the quality of this analysis across different computing platforms.

The approach follows in this work aims to identify the potential and limitations in the use of quantum architectures for the Machine Learning based analysis of simulations data from all atom Molecular Dynamics. Specifically in the field of bimolecular simulations, this analysis is a state of the art and has the potential to provide insightful information on the structure and function of proteins. As the simulations involve large system sizes and can still be performed on classical computers, we aim to question here whether quantum circuits are more efficient and applicable for the analysis of those data instead of classical hardware.