Did you perform multiple simulations of the same complex? That might be required before you venture into error estimations, because a single simulation can be stuck in one of the minima's. Error estimation can be done by calculating auto-correlation times of your data points and then using uncorrelated data points for bootstrap analysis.

Mr. Bhandare,

A statistical data processing involves parameters such as representativeness; accuracy; precision (uncertainty) and more.

Accuracy denotes: "Degree of consistency between the result of determination or the mean of 'n' results and the 'true value' of the analyte measured and determined in a sample"

Or, you should compare theoretical results about energetics with experimental ones. This means a comparative analysis between theoretical and experimental data of a concrete peptide system.

Few experimental methods provide high-precision measurements of energetics, accounting for the fact, that the accuracy of MD computations depends on the accuracy of quantum chemical methods. The accuracy towards energetics of a large part quantum chemical methods is lower than 0.0037 a.u. (or 0.1 eV). DSC, for example, operates with precision +/-0.01 K. The high-precision DSC instrumentation operates with +/-0.0001 K. The proteins exhibit flexible chain, and thus it can occur set conformations with close energetics, comparable to the uncertainity of DSC. There are energy differences ca. (1.10-6)-(1.10-8) a.u. Even operating with precision +/-0.000001 K it could be unable to distinguish fluctuation of the energetics due to small perturbation of the geometery/bonding network. MS, using ICR and Orbitrap analyzers can provide higher precision. The proteins, however, as analytical objects show a complex MS pattern which difficult its quantification, in general. The involvement of CID method allows an exact quantification, however, operating with a lower precision.

This means that, depending on the method, the perturbation of the energetics could prevent to determine the accuracy of your set computational methods, comparing with experimental absolute data, because of the uncertainty of the methods (experimental) can be large for proteins. The uncertainity of the theoretical methods - also. Because of, 'precision' denotes "consistency between data obtained by the same method". For example, you have used BOMD at X (ab initio level) level of theory, to given protein. If you have computed one and the same starting geometry, the precision is high. It is high even employing low level of accuracy of the method X.

But, the manual treatment of a 3D structure, within the framework of subsequently repeated procedures for preparation of the input files (3-5 times) for the computations leads to discrepancy of the energetics, due to fluctuation of disposition of molecular fragment/s. This discrepancy affects the 'reproducibility' of the computed data. The lower reproducibility of your data should be a result of the manual treatment. It is not associated to the accuracy of the method X. In order to exclude such error sources, there are needed single crystal X-ray coordinates as input-file.

However, crystallographic data about proteins are scarce. There is lack of a statistically representative data, allowing to determine experimental uncertainty of geometry parameters of proteins. There are scarce studies devoted to quantification of proteins by spectroscopic/spectrometric methods; and thermodynamics of proteins as well as, comparing with reports devoted to analysis of low molecular weight objects.

In general, the error source shown in the above two paragraphs are associated with docking analysis, where presumably you have operated with known structures of fragment/s of the protein or with a whole protein structure. Because of, within the framework of a homology modeling, there are another set errors due to:

(A) Side chain packing;

(B) Shift of a region, correctly assigned with a template;

(C) Distortion of molecular sequences which have not equivalent in your template protein;

(D) Distortion of a region assigned wrongly in the template; and

(E) Employment of wrong protein template, respectively.

Accounting, however for the fact that, your so-called template, you approximate as true one as 3D structure. But this is only approximation which itself is an error source for your statistics. In this context it is unable statistical analysis with a reference molecule. The error contribution even could not be classified as fluctuation. The reference should has absolutely determined molecular structure. In homology modeling study you cannot apply statistics in general terms. Analytical standard, in the statistics means chemical compound with determined molecular structure (absolutely determined). Because of the properties of this molecule (small or protein) is accepted as the so-called "true" values, towards which ones you have determined the accuracy of your quantities.

You could perform a representative chemometrics using small molecules (ligands to your docking analysis, for example) comparing MS and/or DSC data. This is carried out in order to determine the accuracy of your computational strategy as methods used. The MS profile shows relatively low number fragment peaks, allowing an absolute quantification (position, intensity, energetics) and assignment of the structure of the fragment ions. You could use IR-spectroscopy, too. However the integration accuracy of theoretical method/s together with the integration accuracy of the spectroscopic patterns should be taken into account for.

The use of small molecules for validation of given theoretical quantum chemical computational strategy has set advantages. First, they act as 'references'. There are published large number experimental structural, spectroscopic and/or thermodynamic data allowing their statistical processing. Because of, the term 'representativeness' means “consistency of the data towards a reference material”.

Prior to compute, you can try to find someone/s with qualification analytical chemistry - chemometrics, in order to clarify the representativeness of your conputations in statistical terms. There are needed:

(I) 'Theoretical design' towards your quantum chemical computations, involving:

(i) representative number of protein amino acid sequence (protein fragments of the large protein);

(ii) their complexity towards H-bondingl p-stacking effects and more;

(iii) number of protein/peptide objects (selected set structurally similar peptides/proteins);

(iv) representative set theoretical methods with different accuracy already testes over different chemical clases of molecules; and more;

(II) 'Experimental design' towards measurements of representative analytes (meaning phase analysis, T gradient, P gradient and mode) depending on the properties of your 'reference' molecules; and

(III) 'Theoretical design' towards chemometric methods for statistical data processing of your theoretical and experimental data set/s.

Accordingly, you need contact with experimental chemists about spectroscopic/spectrometric/crystallographic/thermal measurements. Furthermore you can obtain representative quantitative experimental information using non conventional techniques. These measurement conditions you should model theoretically, in order to compare theoretical and experimental results.

Points I–III belong to 'Method validation'. The latter covers topic in Analytical chemistry – chemometrics, reflecting 'reliability' of model/data/conclusions in the chemistry towards real chemical experiment/s.

The validation of theoretical quantum chemistry is associated with uncertainty of experimental data and, in parallel, is connected with uncertainty of quantum chemical results towards predicted energetics due to different effects. The fluctuations are independently accounted for experimental methods and theoretical methods in their own right. These 'uncertainties' should be taken into account for, evidencing conclusions about proteins obtained on the base on theoretical computations even using ab initio computations. Peptides are complex analytical objects both employing theoretical and experimental methods. So, it is not trivial work to evidence computational data of peptides, and a reliable experimental design of 'reference' molecules (point II) is conditio-sine-qua-non.