dpamber

Some useful tools related to Amber and DP.

Installation

pip install dpamber

Tools

corr: generating data for DPRc models

corr tool generates DeePMD-kit training data for DPRc from AMBER sander low-level QM/MM data and high-level data. For details of DPRc, read the DPRc paper.

Before using this tool, one need to prepare low-level and high-level QM/MM data:

$$ E_\text{hl}(\mathbf R)=E_\text{hl,QM}(\mathbf R)+E_\text{hl,QM/MM}(\mathbf R)+E_\text{MM}(\mathbf R) $$

$$ E_\text{ll}(\mathbf R)=E_\text{ll,QM}(\mathbf R)+E_\text{ll,QM/MM}(\mathbf R)+E_\text{MM}(\mathbf R) $$

Low-level and high-level data should use the same coordinate and the same MM method, but different QM methods. So, the correction energy for training will be

$$ \Delta E (\mathbf R) = E_\text{hl}(\mathbf R) - E_\text{ll}(\mathbf R) = (E_\text{hl,QM}(\mathbf R) - E_\text{ll,QM}(\mathbf R)) + (E_\text{hl,QM/MM}(\mathbf R) - E_\text{ll,QM/MM}(\mathbf R)) $$

An example of the command is

dpamber corr --cutoff 6. --qm_region ":1" --parm7_file some_param.param7 --nc some_coord.nc --hl high_level --ll low_level --out dataset

where --cutoff takes cutoff radius of the QM/MM interaction for training. --qm_region takes AMBER mask format for the QM region. --parm7_file and --nc take the PARM7 file and the trajectory (NetCDF) file, respectively. --ll and --hl are the prefixes of low-level and high-level files, including the mdout file (.mdout), the mden file (.mden) and the mdfrc file (.mdfrc). The output dataset directory should be put in --out.

See details from dpamber corr -h.

devi: calculate model deviation

devi can be used to calculate the model deviation of a given trajectory. You need to install DeePMD-kit using

pip install dpamber[dpgpu]

See dpamber devi -h for details.