Simplify Chemical System

[MBartkowiakSTFC]

Description of work
Most of the classes derived from ChemicalEntity were only used for a very specific case of a trajectory created by GROMACS with a PDB file describing the topology. Since MDANSE is moving to MDAnalysis for loading GROMACS trajectories, the selection mechanism based on fixed databases needs to be replaced with a more general solution. This is provided by rdkit and substructure matching. Since rdkit introduces its own indexing of atoms, not guaranteed to be consistent with the internal indexing of the ChemicalSystem class, changes are necessary to improve the consistency of the atom indexing.

closes #545

Fixes

1. ChemicalEntity and all derived classes have been removed from the code.
2. The only database remaining in the Chemistry section is the atoms database.
3. Atom indexing is now fixed and matches the indexing in the coordinate array.
4. New ChemicalSystem implementation stores lists of indices.
5. ChemicalSystem does not have a "configuration" attribute anymore. Configuration is accessed only via the Trajectory class.
6. A backup ChemicalSystem loader allows the users to load old trajectories. The new ChemicalSystem instance will be created in memory and populated with information from the old /chemical_system data structure.
7. The trajectory stores a subset of the atom database, preserving the user-made changes to atom definitions and making trajectories portable even if they use artificial atoms.
8. Atom selection works for trajectories independent of the molecular dynamics engine used to run the simulation. However, it is necessary to detect bonds in the system using the TrajectoryEditor job before it is possible to select structures.
9. Most of the Cython extensions have been replaced with Python equivalents.

To test
All unit tests must pass.
A detailed comparison of the results between this and main branches is necessary due to the fundamental changes to the way the code operates.
Please run TrajectoryEditor with molecule detection on a trajectory and check if all the expected Atom Selection options are working correctly in the GUI.

Jobs that require the most attention:

• CenterOfMassTrajectory
• Voronoi
• SolventAccessibleSurface
• PairDistributionFunction
• VanHoveSelf
• VanHoveDistinct

[ChiCheng45]

I'm going through the code and program. I'll keep posting any issues I find.

I found a bonding issue in the 3D viewer, bonds are being created between the dummy atoms and non-dummy atoms.

[ChiCheng45]

I've been testing the solvent SolventAccessibleSurface job. According to the MDTraj documentation https://mdtraj.org/1.9.4/examples/solvent-accessible-surface-area.html they also use the method by Shrake and Rupley. Here I run the SolventAccessibleSurface with a trajectory from SlimMD https://www.ccpbiosim.ac.uk/slim-md/slimmd-database/proteins/cyclosporin-a-37ms.

Here are the results for the first 1000 time steps.

Protos.

This Branch.

MDTraj.

Concentrating on the timestep at 550 (top) and 670 (the peak, bottom) we can see the molecule geometry change.

At 670 the molecule is in a more open geometry which agrees with the increase in the solvent accessible surface in MDTraj and this branch. Looks like there were some issues with the implementation of the SolventAccessibleSurface on protos.

[ChiCheng45]

Since the extension are all gone we can also delete the MDANSE/Src/MDANSE/Extensions directory.

[ChiCheng45]

I've tested the pair distribution function and the self and distinct part of the van Hove function. Here are the timing with the dftb water trajectory, I think we should try to speed up those numpy functions some more if possible.

Before
pair distribution function: 0.7s
van Hove function self: 37s
van Hove function distinct: 20s

After
pair distribution function: 4s
van Hove function self: 33s
van Hove function distinct: 1m 55s

The self part of the van Hove function looks good but I see differences for the distinct part. Looks like the intra and intermolecular parts haven't been split up.

Here is the the inter and intra PDF for the dftb water trajectory with this branch.

Here is the one on protos.

It looks like the bonding information got lost when the DFTB trajectory was converted.