Molecular simulation derived/inspired

Energy grid histogram

In molecular simulations, the interactions between atoms are described using intermolecular potentials such as the Lennard-Jones potential. Since this potential must be frequently evaluated one can save computational cost by pre-computing the potentials on a grid. The grids themselves are not necessarily fixed-length, and typically high-dimensional. Therefore, they are not directly used as descriptors. Again, one can solve this problem by “summarizing” the grid in form of a histogram.

Converting MOF structures into energy grids and using them as descriptors in form of histograms. Figure taken from [Bucior2019].

Featurizers: EnergyGridHistogram EnergyGridHistogram

considers_geometry: True considers_structure_graph: False encodes_chemistry: True scalar: False scope: global

The feature was proposed in [Bucior2019]. We use the implementation in [Dubbeldam]

Henry coefficent

The Henry coefficient is strongly connected to the energy grid and might sometimes be a more useful (since it is a scalar) descriptor.

Featurizers: Henry Henry

considers_geometry: True considers_structure_graph: False encodes_chemistry: True scalar: False scope: global

We use the implementation in [Dubbeldam]