MD Simulations: Analysis and Ideas

The purpose of this document is to briefly describe how to analyze an MD simulation in layman’s terms. It also offers general simulation ideas.

Analysis Techniques

Root Mean Square Deviation

How similar is the shape of each simulated protein conformation to some reference conformation? The conformation of the first simulation frame is a good choice for the reference. The RMSD should start to stabilize after you’ve been simulating for a while, indicating that the system is properly equilibrated. (Link)

Root Mean Square Fluctuation

How wiggly is each protein residue (e.g., amino acid) over the course of the simulation? (Link)

Clustering

By eliminating protein conformations that are very similar, clustering generates an ensemble of particularly distinct conformations. (Link)

FTMAP Hotspot Analysis

You can use FTMAP to identify druggable hotspots. One approach is to apply FTMAP to representative structures identified through clustering, to account for protein flexibility.

Relaxed-Complex Scheme

Docking small-molecules into various conformations extracted from the simulation via clustering is also a good approach for identifying novel chemical probes. This kind of virtual screen is called the “relaxed complex scheme.”

Ensemble Electrostatics

The electrostatic potentials surrounding the protein can determine how some small molecules bind. You can also calculate ensemble-averaged versions of those potentials that may be more realistic. APBS and Delphi are two programs for calculating potentials.

Principal Component Analysis

Protein motions are very complex. PCA presents a simplified representation of these motions. Some of the minor motions are lost, but the larger-scale motions are still represented. You can project simulated conformations onto the first two principal components (2D graph of the complex motions), or you can morph a model of the protein itself according to the principal components. (Link)

Distance-Based Measurements

It can be helpful to measure the distance between two atoms over the course of a simulation. For example, you can monitor the distance between carboxylate and amine groups to see hydrogen bonds forming and breaking.

Measuring Pocket Volumes

Using the POVME algorithm, you can measure the volume of a given pocket over the course of the trajectory. Sometimes the largest-volume pocket can be useful for drug-discovery projects. POVME is also good at identifying cryptic pockets.

Analyze Hydrogen-Bond Networks

What somewhat-distant residues might influence the motions of the binding pocket through hydrogen-bond networks? HBonanza is a tool for measuring just that.

Pathways of Correlated Motions

You can also analyze the pathways of correlated motions that might connect distant residues. This can sometimes reveal allsoteric mechanisms. WISP is the tool to use.

Kinds of MD

There are endless flavors of MD. I’m going to put a list here, in case you want to research further. (If you do, please feel free to send brief summaries of each method so I can paste them here!)

  • Regular (vanilla) MD.
  • Accelerated MD (McCammon group)
  • Coarse-grained MD (awesome, though I haven’t done much with it)
  • Metadynamics
  • Replica exchange MD
  • Umbrella sampling
  • Markov-State-Model guided MD (amazing)
  • WESTPA (would like to start using)
  • Implicit-solvent MD
  • Brownian dynamics (not really MD, but relevant)

Please send additional methods and/or descriptions if you get a chance. Thanks!

Reasons to Simulate

“Could there be cryptic binding pockets in my protein that aren’t evident in any crystal structure?”

“I’d like to use multiple protein conformations for drug discovery, but all the crystal structures look the same. I’ll simulate to get more conformations!”

“I want to use something better than a docking score to predict ligand binding. Why not use an alchemical method like free-energy purturbation or thermodynamic integration?”

“Could understanding the dnyamics of a certain region of the protein reveal its molecular mechanism?”

“I’ve got two similar proteins, but they do different things. Maybe they are evolutionarily related, or one is a mutant of the other. Can dynamics reveal why they behave differently?”

“My protein has some crazy allosteric mechanism. Might MD simulations reveal the subtle shifts in correlated residue motions that transmit the allosteric signal?”

“I want to engineer a protein to do something new. Can I predict how mutations will change its function before I make the actual protein and test it experimentally?”

“I think I know how a small molecule binds to my protein, but I’m not sure. If I simulate it, will it slip in the binding pocket (probably the wrong pose!), or will its pose remain stable?”

Please send more ideas! I want this page to help with future brainstormig.

Reasons not to Simulate

“I want to see some large conformational changes.” It ain’t gong to happen on MD timescales.

“I want to know something for certain.” You always need experimental validation. You can test something from the literature (already experimentally demonstrated), or you can get a collaborator for prospective validation.

“I want to simulate some huge system.” Probably not going to happen, unless you can get a PRAC.

Leave a Reply

Your email address will not be published. Required fields are marked *