Input files, known as mdin files, set up the conditions that the system is subjected to. Typically, molecular dynamics simulations go through the following steps:

  1. Minimization. Since PDBs do not have hydrogens, and solvent molecules are just kind of thrown into a box by LEaP, the goal of minimization is to, for lack of a better work, minimize the number of clashes and bad contacts between things in the system. Let’s say that you have two hydrogen atoms driving wild protein-based cars. If LEaP decided that the two hydrogen atom cars should be bumper-to-bumper, minimization would be the driving instructor pointedly reminding them to maintain a 3 second following distance. With each step of minimization, you typically lower any restraints that you have (it’s a gradual learning process for our student-driver hydrogens), until your final step which would ideally happen without any restraints at all.
  2. Heating. MD simulations are based on trajectories, and those trajectories need to have initial velocities in order to have kinetic energy. The heating step is where these initial velocities are assigned, and thermal energy (cough heat cough) is added to the system. The temperature starts from 0 K and is gradually brought up to the desired temperature (probably 300 K). Usually the restraints come back during heating because most of what is being heated is solvent.
  3. Equilibration. After the system goes from from being motionless to reaching the desired temperature, the system should undergo equilibration. Equilibration is kind of like the system’s time to relax. It goes from being restrained for most of its life to those restraints being lifted and given the opportunity to wiggle freely. Like with minimization, equilibration often has several steps of gradually reducing the restraint weight, until the final equilibration step that has nothing.
  4. Production. The system is finally ready for unrestrained MD! Huzzah! [And there was much rejoicing.] Production is the part of MD simulations that takes the longest time, but it is also the part that people care about. Analyses are based on the data from the production steps, and production can last as long as you want it to. Typically anything less than 100 ns is useless, and a minimum of 1 μs is seen as ideal. How long you run for is entirely dependent on your resources, but the longer you run for, the better the data.
  • It is wise to break up production into smaller chunks, instead of generating one massive 100 ns data file, though, because writing more frequent restarts means that if the computer crashes you don’t have to start all over, and smaller file sizes means that data corruption is less likely. Additionally, many supercomputers have a wall-clock time, and you lose anything that goes over that time–so writing over that limit means you’re wasting both your time and computational resources.

Replicate trajectories should go through each of these 4 steps independently. So, the initial prmtop and inpcrd should be copied into multiple folders and undergo all these steps alone. The reason for this is that your starting structure won’t change, but what contacts are manipulated through minimization might.

Also, it is **_HIGHLY suggested_** that you check the system after you’ve finished equilibration before moving onto production. Sometimes there are bonds that are clearly not right, and checking will save you a lot of time waiting on a system doomed to fail.