DP model for TYK2

This repository contains scripts for the construction workflow of Deep Potential model for tyk2.

• SYSTEMS.txt: name of 15 ligands
• prepare_gaussian_input.py: prepare Gaussian input files from conformers sampled by GROMACS
• gaussian.tar: input and output of Gaussian
• log2dp.py: convert gaussian log file to deepmd format
• lcurve_plot.py: script to visualize learning curve
• plot.json: a sample of settings for plot learning curve：
  • win_length: the window length to smooth learning curve
  • fig: the directory to save figure
  • mode: "trn" or "tst", whether to use training or testing loss in lcurve.out file
  • loglog: true or false, whether to change axis to log scaling
  • lcurves: the lcurve.out files to plot
  • labels: the legends for each file in lcurves

To plot learning curves, just run python lcurve_plot.py plot.json

• test.py: script to evaluate energy and forces for all conformers in training data
• plot_err_distribution.py: script to plot unsigned error distribution and RMSE

ligands_5ns

• ligands_5ns/*/md_traj.gro: trajecory of 5-ns simulation of 15 ligands in solvated phase

data

Training set in DeepMD-kit format

models

• lcurve.png: a comparison between learning curves of models with tanh/gelu activation function
• lcurve_smooth.png: curve without smoothness
• gelu/tanh: contains frozen_model.pb, lcurve.out and input.json files

eval

• */*_e.txt: results of energy prediction
• */*_f.txt: results of force prediction
• error_distribution.png: error distribution plot