Folding of villin miniprotein was studied by parallel tempering metadynamics driven
by machine learning. To obtain a training set for machine learning, we generated
a large series of structures of the protein by the de novo protein structure prediction
package Rosetta. A neural network was trained to approximate the Rosetta score. Parallel
tempering metadynamics driven by this approximated Rosetta score successfully predicted
the native structure and the free energy surface of the studied system.

These files make it possible to rerun all simulations. The directory METAD contains input
files for metadynamics (no folding events observed). The directory PT-METAD contains input
files for parallel tempering metadynamics. All simulations were done using Gromacs 2016.4,
Anncolvar 0.8, Plumed 2.4 and OpenMPI 4.0.0.

METAD/
  justcommands.sh - commands for running the simulation
  md.mdp - Gromacs input file with simulation parameters
  npt.cpt - Gromacs checkpoint file with high-resolution input structure
  npt.gro - Gromacs file with low-resolution input structure
  plumed.dat - Plumed input file with metadynamics options
  ref.pdb - Reference structure for MOLINFO Plumed procedure
  ref2.pdb - Reference structure for FIT_TO_TEMPLATE Plumed procedure
  topol.top - Gromacs topology file

PT-METAD/
  justcommands.sh - commands for running the simulation
  md_0.mdp - md_42.mdp - Gromacs input files with simulation parameters (for different
                         temperatures)
  nvt0.gro - nvt42.cpt - Gromacs checkpoint files with high-resolution input structures
                         (for different temperatures)
  nvt0.gro - nvt42.gro - Gromacs files with low-resolution input structures (for different
                         temperatures)
  plumed.dat - Plumed input file with metadynamics options
  ref.pdb - Reference structure for MOLINFO Plumed procedure
  ref2.pdb - Reference structure for FIT_TO_TEMPLATE Plumed procedure
  topol.top - Gromacs topology file

md5.log - MD5 signatures for files