AI3SD Video: Making sense of highly flexible molecular simulations: Where AI can help and where not

Abstract

With simulating the dynamic behaviour of ever bigger molecular systems for longer simulation time we simultaneously achieve more realistic timescales and gain much better insights into the physiological relevant time dependent behaviour of molecular systems, but also generate significantly more data and thus pose new challenges for filtering noise and analysing the simulation data. In simulation analysis and data dimensionality reduction we often rely on linear dependencies and behaviour within the simulated timespan. This generally is true for systems that show slow structural or conformational transitions over the simulated timespan. For example, proteins and enzyme simulations that undergo large scale conformational changes can adequately be analysed by methods of principle component analysis (PCA) and analysing and visualising low frequency normal modes. However, much more flexible molecular systems undergoing multiple and seemingly chaotic conformational changes are posing challenges for their analysis. Here, we need to advance our analysis toolbox. Moreover, it is important to understand the flexibility and linear behaviour of the simulated system before choosing the analysis methods. In this talk we present three very differently behaving molecular systems including an enzymatic activation process, a flexible self-assembling host guest system, and a highly flexible dataset of lipid molecules relevant for antibiotic resistance of mycobacterium tuberculosis. We will show how chaos theory can help to understand the flexibility patterns of the simulations, then present classical PCA based simulation analysis, before introducing the opportunities and challenges for unsupervised machine learning methodologies. The presented methods will concentrate on competitive learning methods (self-organizing maps) and density based clustering algorithms (DB scan) to analysis dominating and hidden structural features in seemingly chaotic simulation data.

More information

Identifiers

ORCID for Jeremy G. Frey: orcid.org/0000-0003-0842-4302

ORCID for Samantha Kanza: orcid.org/0000-0002-4831-9489

ORCID for Mahesan Niranjan: orcid.org/0000-0001-7021-140X

Catalogue record

Export record

Altmetrics