sungroup-sjtu / AIMS_Simu

High-throughput Molecular Simulation (HMS) workflow server

Molecule Simulation Database -- Server

This project performs high-throughput force field simulation and data processing
This project depends on AIMS_Tools
Two main scripts accomplish the jobs submit.py and monitor.py.

To setup on a Linux server

1. Clone AIMS_Simu and AIMS_Tools:
   It is suggested to put them in same folder.

   git clone https://github.com/sungroup-sjtu/AIMS_Simu  
   git clone https://github.com/sungroup-sjtu/AIMS_Tools
   

2. Modify config.py
   a) Set paths of MS_TOOLS, WORK_DIR, PACKMOL, DFF and DFF database.
   b) Select the job queue (fast, gtx, cpu on Cluster 86)
   c) Set GROMACS executable

   For example:

   force field setting(no default setting):
       DFF_TABLE = 'MGI' # 'IL'
   paths setting:
       MS_TOOLS_DIR = os.path.join(CWD, '..', 'AIMS_Tools') # AIMS_Simu and AIMS_Tools in same folder
       WORK_DIR = os.path.join(CWD, 'SimulationData') # all simulation data is saved in a new folder AIMS_Simu/SimulationData
       PACKMOL_BIN = '/share/apps/tools/packmol'
       DFF_ROOT = '/share/workspace/xiangyan/src/DFF/Developing' # simulation paramters come from this folder
   PBS settings:
       ...
       PBS_ARGS = ('gtx', 32, 2, 16)  # partition, cpu, gpu, cpu_request
       ...
       GMX_BIN = '/share/apps/gromacs/2018.6/bin/gmx_serial'
       GMX_MDRUN = 'gmx_gpu mdrun'
       # GMX_MDRUN= 'gmx_fast mdrun'
       GMX_MULTI = True
       GMX_MULTI_NJOB = 8  # Use -multidir function of GROMACS. For Npt simulation, set it to 8. For NvtSlab simulation, 4 is better
       GMX_MULTI_NOMP = None  # Set the OpenMP threads. When set to None, use only one node and the best number of threads is automatically determined
   simulation details settings (default setting is OK):
       NATOMS = 3000 # least number of atoms build in simulation box.
       NMOLS = 120 # least number of molecules build in simulation box.
       LJ96 = False # using LJ 9-6 non-bonded potential
       DIFF_GK = False # using green-kubo method to calculate the diffusion constant. (Expensive, not suggest)
       DEBUG = False # if true: do not delete the trajectory file in analyze process.
       class NvtMultiConfig(Config, SunRunConfig, SunExtendConfig, SunBugFixConfig):
           REPEAT_NUMBER = 80 # set the number of parallel simulation for nvt-multi
   

3. To set up a high-throughput computation. Prepare a list of molecules in /mols/ and then:
   the example.txt contains 5 columns: name SMILES molecular_ratio t_list p_list
   more than 5 points for t_list and p_list is needed, otherwise some analysis and dumps scripts will not work.
   run/submit.py -p [npt,nvt-slab] -i mols/example.txt -r 'comments' -tp assigned

4. To run the calculations:
   run/monitor.py -p [npt, nvt-slab]

   Available procedures: npt, nvt-slab, ppm(npt), nvt-multi(npt).
   ppm(npt) means npt is prerequisite of ppm.
   Example for npt procedure:

   cd run
   ./submit.py -p npt -i mols/example.txt -r testing -tp assigned
   ./monitor.py -p npt 
   

5. The results are saved in the WORK_DIR. In default, WORK_DIR=AIMS_Simu/SimulationData.

QM calculation

QM calculation for heat capacity is performed by run-cv.py script

1. Prepare QM files. This will check database to remove duplicated molecules from example.txt and process the molecule name. A file named _cv_prepared.txt will be generated and used for following steps
   ./run-cv.py prepare mols/example.txt
2. Generate Gauss input files and submit to PBS job manager
   ./run-cv.py cv _cv_prepared.txt
3. Analyze Gauss results. The results will be saved in a file named _cv.log
./run-cv.py get-cv _cv_prepared.txt
4. Save results into database
   ./run-cv.py save-db

For more information, see our publication: "Predicting Thermodynamic Properties of Alkanes by High-throughput Force Field Simulation and Machine Learning", https://doi.org/10.1021/acs.jcim.8b00407

Scripts for data post-processing and analyzing

Several scripts are provided for post-processing and analysing the simulation data. They are located at scripts and scripts-post

1. Fitting the simulation data at different temperature and pressure. So that properties and derivatives at arbitrary T or P can be obtained.
   This should be performed prior to any other analyzing
   ./scripts/post-process.py -p [npt, nvt-slab] -o True
2. Remark molecules containing specific groups (e.g. halide, cyclo-ester) as bad molecules, which will not be dumped in following steps
   ./scripts/remark.py [npt, nvt-slab]
3. Dump the molecules from sqlite database to mols.csv file. The category should be specified, which is necessary for uploading to AIMS_Web database
   ./scripts/dummp-mols.py [small molecule, ionic liquid, ...]
4. Dump the simulation data from sqlite database to csv file, which can be uploaded into AIMS_Web database
   ./scripts/dummp-data-npt.py
./scripts/dummp-data-nvt-slab.py
5. You select specific class of molecules in following analysis based on force field atom type, by modify the app/selection.py.
6. Compare with NIST Experimental data
   • Make sure that nist.sqlite exists in database folder
   • Run following script to compare simulation and expt data and plot the results

     cd scrips-post
     python3 compare_detail.py -p npt -t nist --selection False
     python3 compare-nist.py -p npt --selection False
     

7. Compare with ILTHERMO Experimental data
   • Make sure that ilthermo.sqlite exists in database folder
   • Run following script to compare simulation and expt data and plot the results

     cd scrips-post
     python3 compare_detail.py -p npt -t ilthermo --selection False
     python3 compare-nist-il.py -p npt --selection False