Molecular dynamics serves as a powerful tool for studying molecular systems, offering researchers insights that are often unattainable through experimental methods. With greater accessibility and numerous free tutorials, more researchers are joining the field each year. However, MD is not a pot where you can mix ingredients and leave them to cook on their own. Minor errors in preparation or interpretation can lead to misleading results that seem valid but are scientifically inaccurate.
For newcomers, the challenge is that many common pitfalls are subtle and often not discussed in basic tutorials. These mistakes may not cause simulation crashes, which makes them easy to overlook until much later. In many cases, a supervisor or referee detects them after months of wasted computational work.
This article highlights ten of the most frequent mistakes made by beginners in MD simulations. By understanding these pitfalls early, new researchers can produce more reliable simulations and avoid the frustration of interpreting flawed or unstable systems.
I have written another article on the basics of molecular dynamics simulation. If you are a new researcher in this field, you might find that article helpful.
1. Poor Preparation of Starting Structures
The quality of a molecular dynamics simulation is directly dependent on the starting structure you provide. If the initial model contains missing atoms, steric clashes, wrong protonation states, or unrealistic geometries, the entire simulation may behave incorrectly. Many new researchers assume that a structure downloaded from an online PDB database is immediately ready for MD, but this is rarely the case.
Proper structure preparation requires checking for missing residues, selecting the correct tautomers, assigning appropriate protonation states at the pH, and correcting any structural gaps or distortions. Skipping these steps may cause the simulation to fail during minimisation or produce unrealistic motions later on.
Fortunately, there are several online tools and software, such as pdbfixer and H++, that can assist you in the preparation step.
2. Using an Incorrect Time Step
Choosing an inappropriate timestep is a frequent mistake by novice molecular dynamics practitioners. If the timestep is too large, the numerical integration may become unstable. Bonds may stretch too far, atoms may move unrealistically fast, and finally, the simulation may blow up. This problem is especially common when beginners use a large timestep without constraining hydrogens or enabling virtual sites.
On the other hand, selecting an unnecessarily small timestep wastes computational resources. While it increases stability, it drastically slows down simulation progress without significantly improving accuracy. Many students use excessively short values because they are uncertain about the system’s requirements.
Finding a suitable timestep requires balancing accuracy and efficiency. The correct value depends on the force field, bonded constraints, masses, and presence of virtual sites. A well-chosen timestep ensures stable dynamics while maximising simulation time from available computational resources.
3. Neglecting The Artefacts Caused by Periodic Boundary Conditions
Periodic boundary conditions (PBCs) are essential for simulating bulk systems, but they also introduce artefacts that must be handled carefully. PBC can cause molecules to appear split across boundaries or diffuse into images of themselves. For example, a ligand may appear to jump suddenly, or a protein might look distorted simply because it crossed a periodic boundary. While these behaviours are not simulation problems, they can complicate analysis if not corrected.
Many structural analyses, such as RMSD, DSSP, distance-based analyses, hydrogen bonding, or clustering, can yield incorrect results if we don’t treat the system’s PBC properly and make the molecules whole before measurement.
Fortunately, most MD software packages include built-in tools for correcting PBC artefacts, making it straightforward to prepare clean and physically meaningful trajectories for analysis. GROMACS offers versatile options in the gmx trjconv command to rebuild broken molecules, remove jumps, and keep the system properly centred. AMBER provides similar functionality through the cpptraj program to handle even complex PBC issues. However, understanding PBCs is essential for avoiding misleading conclusions.
4. Inadequate Minimization and Equilibration
Many beginners rush through the minimisation and equilibration steps because they want to begin production runs quickly. However, minimisation removes bad contacts and relaxes high-energy regions of the system, which prevents instabilities during MD. Skipping or shortening this stage increases the risk of structural distortions or catastrophic simulation crashes.
Equilibration is equally essential because it allows temperature, pressure, and density to stabilise before production begins. If these quantities are not properly equilibrated, the system may not represent the correct thermodynamic ensemble. As a result, measurements such as diffusion, binding, or conformational stability are unreliable.
As an in-silico researcher, you should always ensure that minimisation converged and the system has reached an actual equilibration state by verifying that key thermodynamic properties, such as temperature, pressure, total and potential energy, and system density, have stabilised and formed consistent plateaus under the chosen ensemble conditions.
5. Using an Unsuitable Force Field
Force fields are carefully designed and parameterised for specific classes of molecules, such as proteins, nucleic acids, membranes, or small organic ligands. A common mistake is to choose a force field simply because it is widely used, without considering whether it is appropriate for the particular system. Using the wrong force field leads to inaccurate energetics, incorrect conformations, or unstable dynamics.
Applying a protein-specific force field to carbohydrates, or using an outdated version when newer, more accurate models exist, can introduce serious errors. For instance, CHARMM36m is designed for proteins, UFF4MOF is tailored for MOFs, and GAFF2 is a general-purpose force field for organic molecules.
The best approach is to select a force field that has been developed for the specific type of system under study. Reviewing literature and community best practices ensures that your chosen model is accurate and widely accepted. Selecting the correct force field results in more realistic trajectories and more reliable outcomes.
6. Mixing Incompatible Force Fields
Combining parameters from different force fields can result in inconsistent or unphysical interactions. This issue arises because force fields employ different functional forms, methods for deriving charges, and Lennard-Jones combination rules. When these elements are merged carelessly, it disrupts the balance between bonded and non-bonded interactions, leading to unrealistic behaviour in simulations.
Beginners may attempt to mix force fields when dealing with complex systems that include different parts, especially when the main force field cannot handle some molecules in the system. However, unless the individual parameter sets are explicitly designed to work together, mixing force fields can lead to disastrous outcomes.
When a custom component, such as a ligand, is needed, it is essential to generate parameters that match the main force field. Some force fields are designed to work together, such as CGenFF with CHARMM36 or GAFF2 with AMBER ff14SB. Therefore, before mixing force fields, ensure they are compatible.
Consistency ensures that all system components interact under the same physical assumptions, which is crucial for reliable molecular dynamics results.
7. Skipping Proper System Validation
Many new researchers assume that if a simulation runs without crashing, it must be correct. This assumption is misleading because MD engines will happily simulate a system even when key components, such as protonation states, force field parameters, or bonded interactions, are incorrect. Validation before production runs is essential to confirm that the setups accurately reflect the system’s physical reality.
Proper validation can include several key checks, such as verifying energy values, examining diffusion coefficients or viscosity, ensuring stable temperature and pressure, visually inspecting the system, and, most importantly, calculating simple observables and comparing them with experimental data or a higher-level model that may be physically more accurate.
Validating a molecular dynamics simulation using experimental data typically involves comparing the observables derived from simulations with structural, dynamic, and thermodynamic properties observed experimentally. For example, we can relate experimental X-ray crystallography data, such as B-factors and atomic coordinates, to RMSF and RMSD obtained from simulations, or we can compare experimental NMR observable data, such as Nuclear Overhauser Effect (NOE) distances and scalar coupling constants, with their simulated counterparts.
To learn more about the validation process, read this, this and this paper.
8. Misinterpretation of Metrics and Equilibration Indicators
Beginners often rely on basic metrics, such as RMSD, to assess whether a system is stable or has reached equilibrium. However, a flat RMSD curve does not necessarily indicate proper thermodynamic behaviour or confirm that the system is in equilibrium. Significant information may be overlooked if we focus only on RMSD, as other properties, such as energy fluctuations, radius of gyration, hydrogen bond networks, and diffusion behaviour, can provide vital insights.
Misinterpreting metrics can lead to incorrect scientific conclusions. For example, a ligand may appear stable in the binding site based solely on RMSD, while in reality, hydrogen bonds are breaking, pocket volume is changing, or the ligand is preparing to dissociate. Similarly, a protein may show stable RMSD but still have unresolved secondary structure transitions.
Proper evaluation involves combining multiple observables, reviewing the expectations in the literature, and understanding how each metric reflects the underlying physics. Effective analysis is multidimensional and should not rely on a single curve or number. This approach helps avoid the mistake of drawing conclusions from incomplete evidence.
9. Insufficient Sampling and Lack of Replicate Runs
New researchers in MD often mistakenly believe that a single trajectory represents the complete behaviour of a molecular system. So they stop after running a short simulation and assume the results represent the system’s true thermodynamics. In reality, to obtain scientifically valid, reproducible, and statistically meaningful results, it’s essential to run an MD simulation multiple times.
There are several reasons for this necessity. First, while molecular dynamics is deterministic in its fundamental equations of motion—meaning that a given set of initial conditions, combined with a specific algorithm, uniquely determines the system’s future trajectory—practical factors introduce nondeterminism. In long simulations, small errors can accumulate, leading to outputs that appear random. As a result, there is no guarantee that two different runs will produce the same results.
Another issue is that a single trajectory rarely captures all relevant conformations. This is particularly true for biological systems, which have vast conformational spaces that MD must overcome many energy barriers to explore significant conformational states. Therefore, it is necessary to conduct multiple molecular dynamics simulations using different initial velocities.
A single simulation may fall into a local minimum or follow a particular pathway that is not statistically representative. Running multiple independent simulations provides a clearer picture of natural fluctuations and increases confidence in observed behaviours. Without replicates, observed processes may merely be noise or artefacts.
10. Insufficient Attention to Simulation Parameters
Simulation parameters, such as thermostat type, barostat settings, cutoffs, constraints, and integration algorithms, directly determine the simulation’s physics. New researchers often use default or tutorial settings without understanding their meaning. Small mistakes here can lead to incorrect ensembles, unrealistic pressure or temperature behaviour, or numerical instabilities.
As a competent in silico researcher, you should never use a simulation parameter without knowing its meaning and its effect on the simulation. Always check different values for a simulation parameter and select the one that best suits your system and study. For example, a tutorial may use the steepest descent integrator to minimise the system; however, this method may not converge on your system, and you must know how to switch to a more robust algorithm, such as the conjugate gradient. Or you may need to use a higher interval value when updating the neighbour list due to the specific hardware you use.
All molecular dynamics programs include a detailed manual outlining the program’s parameters. For instance, you can find a complete list of GROMACS simulation parameters available online.
Paying close attention to the simulation settings is essential to ensure that the molecular dynamics simulation correctly reflects the appropriate thermodynamic conditions and that the results can be trusted. Additionally, understanding the simulation parameters helps you recognise both the conditions and the limitations of the simulation results.
If you have experience in molecular dynamics and can share valuable insights, practical tips, or examples of common pitfalls, I would greatly appreciate your contribution. It could help others avoid costly mistakes and improve the quality of their simulations.
I also have another article on the basics of molecular dynamics simulation. It would be my pleasure if you had a look at it.