QUBEKit2Testing

All relevant information and input files for the testing of QUBEKit 2

QUBEKit config files, pdbs and xmls are all in the respective run folder, as per spreadsheet

Each folder for training and testing contains:

• Final pdbs and xmls for each molecule.
• A summary of results from QUBEBench.
• The output file from ForceBalance (optimise.out).
• A QUBEKit input file which can be used to generate the same data (.json file).

Optimising and Evaluating Rfree Parameters

In order to optimise a set of Rfree parameters for a particular training set, ForceBalance requires a forcefield for all molecules in a directory named forcefield.

Having run QUBEKit and obtained a set of xmls, simply run the xml_combiner.py script in the directory which contains all QUBEKit run directories. This will generate a combined.xml file containing the forcefield for all molecules in the training set. Any virtual sites used will be added and atoms will be appropriately named and numbered. This is the forcefield file required by ForceBalance.

The xml_combiner.py script contains logic to show ForceBalance which parameters to optimise. These are in the Lennard-Jones section of the script, where, for example, epsilon is calculated as follows:

epsilon={bfree}*({vol}/{vfree})/(128*PARM['{ele}Element/{free}free']**6)*{57.65243631675715}

Here, the PARM['{ele}Element/{free}free'] describes which element's Rfree value is being optimised, as outlined in the ForceBalance sub-element of the xml file. More information on selecting custom parameters to optimise can be found in the ForceBalance documentation.

---

Next, ForceBalance requires a target folder, containing files for each molecule:

• A file containing the molecule coordinates used for gas simulations, such as a pdb file.
• A file containing molecule coordinates of several (500) molecules, to be used in liquid simulations.
• A data file, which shows information such as the simulation temperature, expected empirical results, and weighting if some empirical parameters are more important than others.

For the targets, simply generate a standard liquid box with the desired number of molecules, and edit a data.csv file to contain the relevant values.

These liquid boxes can be generated using QUBEBench. QUBEBench doesn't need to be installed, it can be run as a script. Simply run python qubebench/run.py -csv <name>.csv to create a csv with all molecules in the current working directory. QUBEBench can then read this csv to either analyse the liquid properties:

python qubebench/run.py -bulk <name>.csv

Or generate the relevant files for ForceBalance with -fb:

python qubebench/run.py -fb true

This command will create liquid boxes of the appropriate size, as well as generating blank data.csv files, gas pdbs and all subfolders needed by ForceBalance.

An example data.csv file is laid out here, with some further example files in the targets folder of this repository:

Global	rho_denom	30
Global	hvap_denom	3
Global	kappa_denom	5
Global	cp_denom	2
Global	eps0_denom	2
Global	use_cvib_intra	FALSE
Global	use_cvib_inter	FALSE
Global	use_cni	FALSE
T	P	MBAR	Rho	Rho_wt	Hvap	Hvap_wt
298.15	1.0 atm	FALSE	787	1	37.8	1

Finally, an optimise.in run file is used to control some fitting parameters, and point ForceBalance to the correct files.

An example optimise.in file is given in this repository. A file for a single molecule optimisation would be as follows:

$options
forcefield combined.xml
jobtype newton
trust0 0.3
penalty_type L2
penalty_additive 1.0
print_gradient 1
print_hessian 0
print_parameters 1
error_tolerance 1.0
mintrust 0.1
convergence_gradient 0.001
convergence_objective 0.1
convergence_step 0.005
$end

$target
type Liquid_OpenMM
name mol01
liquid_eq_steps 100000
liquid_md_steps 1000000
liquid_timestep 1.0
liquid_interval 1.0
$end

Note the forcefield flag points to the combined.xml file in the forcefield folder, and the name flag points to any molecules in the targets folder.

With the directory structure now like this:

- main
    - optimise.in
    - forcefield
        - combined.xml
    - targets
        - mol01
            - data.csv
            - gas.pdb
            - liquid.pdb

ForceBalance can be run with the command:

ForceBalance optimise.in

The resulting optimise.out file containing the optimised Rfree parameters can be used to produce a new force field. QUBEKit can read this file automatically, or the parameters can be put into QUBEKit's json config file. QUBEKit can then be run as normal for any number of molecules. To ensure best results, the run configuration should be the same as was used before the Rfree parameter optimisation (same basis set, virtual sites, etc). If being read automatically, a message will be displayed to the terminal confirming the Rfree parameters are being used.

"Updated free parameter from ForceBalance file."