Definition of the terms "modeling" and "simulation" in the context of biomolecules and/or proteins

Question

What are the definitions of the terms "modeling" and "simulation" in the context of biomolecules and/or proteins?

Examples will be highly appreciated.

Also, please, back them up with authoritative references.

The definitions I found don't reflect molecular biology or related fields. E.g. Principles of Modeling and Simulation A Multidisciplinary Approach (edited by John A. Sokolowski and Catherine M. Bank) contains those definitions.

MODELS: APPROXIMATIONS OF REAL - WORLD EVENTS

A model is a representation of an event and/or things that is real (a case study) or contrived (a use - case). It can be a representation of an actual system. It can be something used in lieu of the real thing to better understand a certain aspect about that thing. To produce a model you must abstract from reality a description of a vibrant system. The model can depict the system at some point of abstraction or at multiple levels of the abstraction with the goal of representing the system in a mathematically reliable fashion. A simulation is an applied methodology that can describe the behavior of that system using either a mathematical model or a symbolic model [4] . Simply, simulation is the imitation of the operation of a real - world process or system over a period of time [3] . As you will see there are many uses of M & S. M & S can be used to determine the ordering policies of Wal - Mart ’ s extensive inventory system, or it can be used to analyze the prospects and rate of rehabilitation of a patient who just underwent knee - replacement surgery, or it can be used to evaluate ocean currents and waves to better understand weather patterns.

M & S begins with 1) developing computer simulation or a design based on a model of an actual or theoretical physical system, then 2) executing that model on a digital computer, and 3) analyzing the output. Models and the ability to act out with those models is a credible way of understanding the complexity and particulars of a real entity [4] . From these three steps you can see that M & S facilitates the simulation of a system and then a testing of the hypothesis about that system. For example, if you wanted to determine how many cashiers are needed to process a certain number of customers during rush hour with the assurance that the store ’ s high level of quality service was not compromised, you must fi rst research the current system of processing customers.

However, I am not sure if they are applicable for molecular or biomolecular modeling.

Answer 1

You simply cannot go past Allen and Tildesley's Computer Simulation of Liquids as a standard textbook on molecular dynamics. Here is a particularly illuminating diagram:

As the diagram says: (italics for nouns in rounded boxes, bold for verbs in cornered boxes)

• When you have some real liquids, you can perform experiments to obtain some experimental results.
• You can also make models of your real liquids to obtain model liquids.
• When you have some model liquids, you can carry out computer simulations to obtain results for the model liquid. You must compare these results with the experimental results as a test of your models!
• And when you have some model liquids, you can also construct approximate theories to obtain theoretical predictions. You must compare these predictions with the model results as a test of your theories!

---

So, as a protein-specific example, real proteins are experimentally observed to have one stable structure. The theory for this is Anfinsen's dogma. A computer model for a protein would be to describe the protein (and the water and ions around it) as a bunch of particles, each corresponding to an atom. (It would be a computer model because it's impractical to make any predictions from it without a computer.)

Two essential ingredients for the model would be the dynamics integration of the particles -- usually Newtonian dynamics with thermostatting, the latter of which is not trivial, and the algorithm by which you evolve the equations with a finite timestep -- and the force fields that say what forces go into the Newtonian dynamics, as a function of the particles' positions and other input.

From a bunch of (statistically independent) initial positions, we would then run simulations of the model and see what final states we get. A realistic protein force field will predict that the protein folds up into one well-defined state. If we didn't see that in our simulations, we might have a bad model, but we might also have not run our simulations for long enough to get good statistics.

---

As such, ideally, well-run simulations of an informative model would give us valuable insight into a real-world phenomenon (in the example, helping us understand the molecular mechanisms behind protein folding). But well-run simulations will not help us if our model is not informative (the bane of all computational chemists!), for example if it is too simple, or biased, or too complex to make useful inferences. Or we might have an informative model, but poorly-run simulations will not help us (if the program has bugs, or our computers are not fast enough to run large enough simulations).

Or, worse still, you might have a poorly-run simulation of an uninformative model -- in which case you will not learn anything about the real world. (Even if the work gets published, which is sadly still quite likely!)