It depends on the use you make of such comparison. In CASP, to compare protein model predictions with the experimental structure, other parametrs are evaluate. I think that two isoforms with RMSD 0.5 have a very similar global structure, with differences in crucial regions. And you should see at these regions in detail.

Matteo,

I'd recommend taking a look at THESEUS (freely available at, www.theseus3d.org). It allows for super-positioning given a sequence alignment, which may be useful when considering different isoforms of a protein (see, Theobald & Steindel, Bioinformatics (2012) 28(15):1972-1979). In addition to providing an over-all RMSD (and other RMSD like measures) it can output a wealth of useful diagnostics, such as residue-wise residuals, which may help you pinpoint specific structural deviations. Alternatively, it may help you assert that the structural deviation is evenly distributed across the entire sequence.

you can do a dihedral angle PCA on the two structures, that should tell you to what extent they are different. Dihedral angles being internal coordinates do not suffer from the problems of fitting/superposing the two structures.

Well the RMSD is indeed a tricky measure however it depends on what you are superimposing, specially depending on the respective size because the RMSD is not normalized. If you use PyMOL you can create your own selection to be superimposed instead of just using the align command. Otherwise there are alternative for pairwise alignment such as Dali (http://ekhidna.biocenter.helsinki.fi/dali_server/) and TMalign. Both will optimize the structural superimposition and then return you the correspondign sequence alignment, the RMSD and in the case of TMalign a TM score (http://zhanglab.ccmb.med.umich.edu/TM-align/). Both are good and will surely deliver you a you results, specially considering you have really close structures. PS you don't have to install the software, you can use the webservers.

Rely on First pairwise sequence alignment, then perform structure superimposition, Yes, RMSD may only assist you numerically, you have to look upon the aligned structures using a visualizer or a structure manipulation program to identify the regions of similarity and disordered domains

It seems the proteins are very similar in your case. Thus, superimposing them on the similar elements of the secondary structure would give rather low RMSD. In this case this is a good and precise characteristic of the similarity.

As Angelo and Julian say, it depends on what you want to use the comparison for. If you have a small region of the structure that is of high interest, such as an active site cavity, then to isolate any differences in this cavity between the two structures, you need to superimpose the proteins using some or all of the residues surrounding this cavity.

Once superimposed, you can use cavity analysis tools such as GRID or SuperStar to compare the possible binding properties of the cavity. There are also some nice secondary structure decomposition tools that could be used to isolate differences in turns/helices/sheets between the two structures. SecBase that is part of Relibase+ from CCDC is a particularly comprehensive secondary assignment method.

Use TM-align:

http://zhanglab.ccmb.med.umich.edu/TM-align/

If sequences are the same use TM-score:

http://zhanglab.ccmb.med.umich.edu/TM-score/

You can use STRAP align for both RMSD and TM-score.

The answer set you've accrued thus far has an interesting diversity of valid and potentially useful answers. I think it would be helpful to those giving advice if Dr. Marangon could provide a little contextual basis. Specifically, what is the practical issue at stake? One might look to similarity analysis as a means for ranking multiple proteins according to likelihood of conserved structure and function; one might want to identify areas of conservation and divergence; perhaps there is some other objective. The selection of specific method may well vary as a function of exactly what one would like to do.

But regardless of the objective, it may be useful to look beyond the predicted static structure and see if there might be important dynamic differences between the two structures. Short of doing a rigorous molecular dynamics simulation, the best way to address this is through a normal mode analysis server such as El Nemo (http://www.igs.cnrs-mrs.fr/elnemo/).

I recommend using Pymol for visualisation and alignment. Load both PDB files into the program and use the align function. The program will report the RMSD (root mean square deviation) among C-alpha carbon atoms. See www. pymol.org

when looking at such similar proteins as isoforms RMS maybe an insensative measure as this is usually done on Calpha coordinates which should be almost identical for isoforms. You are really interested in side chain differences or very local deviations of amino acids. A single number measure of similarity is not going to mean much so there is no way arond spending a significant time with a graphics system looking for the side chain differences. It will help to have very good superimposition of the molecules if you are going to look at them. The most thurough program I find is Rapido http://webapps.embl-hamburg.de/rapido/ which checks if there any labile domains and treats them separatey snd produces many different superimposition models.

You can superimpose two proteins which you think are similar and compare them based on the RMSD values. You can do this any visualization tool such as pymol.. SPDBV or using online tool such as superpose

I agree with Miroslav,

As I see almost all protein isoforms have near identical back bone conformation. Therefore, measuring RMSD between backbones gives you nothing interesting. It would be better to check RMSD by all atom or at least side chains not C-alpha alone.

I recommend you using ICM-Pro or browser, which performs several modes of superposition.

There are a lot of similarity measures. I agree with whose who say that it is depends of what you want to discover. At the last ISMB-ECCB in Berlin it was a talk about CMO algorithm and modified distance matrix analysis.

You can use Mammoth, TM Score (for similar structures), ProFit, TopMatch. This programs superimpose two PDB structures and calculates many structurals measures.

Hi all and thanks a lot for all the useful inputs! I really appreciate it!

To answer some of your questions here is some more context on my problem.

The two proteins I have solved have very similar 3D structure, but their sequences are only 70% identical (201 vs 198 residues). However, their folding is really really similar, except for 3 loop regions that differ a bit. This is why I agree with those of you saying that in my case a general RMSD estimation is quite pointless as it is always going to be very low given the similarity of the backbones. Probably I will take a look at RMSD of those loops that differ with the PyMol function and see what happens.

What I am trying to achieve here is to explain by structural comparison the reasons why these two proteins have very different stability and aggregation behavior upon unfolding as shown by DSC and DLS experiments.

AS many of you suggested I tried the TM-align tool and the results were:

0.0 < TM-score < 0.30, random structural similarity

0.5 < TM-score < 1.00, in about the same fold

Aligned length= 196, RMSD= 0.65, Seq_ID=n_identical/n_aligned= 0.709

TM-score= 0.97759 (if normalized by length of Chain_1)

TM-score= 0.96315 (if normalized by length of Chain_2)

(You should use TM-score normalized by length of the reference protein)

Hi Julian,

they are both thaumatin-like protein isolated from the same medium. but they eluted in different conditions during purification and they crystallized in different media. They have basically the same MW, the same function and they are recognized as two types of the same protein when you analysed them by LC/MS-MS. Maybe it would be more appropriate to define them as two very similar proteins belonging to the same protein class.

In this case best case scenario is to use the TM score from Zhang lab...and there are definite reasons for the same...please check the answer provided by Jeffrey Brender in the following thread...https://www.researchgate.net/post/which_is_the_best_software_program_database_to_check_and_compare_overlay_my_protein_3D_structures...this should answer your question as this score takes care of the problems with RMSD measurements

When the sequences are identical you can use TM-Score or RMSD. TM-Score uses multiple subsets of atoms to seed a superposition and avoids some of the problems of RMSD as described in Zhang's paper.

If the sequences are not identical - you must first either manually or automatically determine which set of atoms are the equivalent ones between the two proteins and then superpose those atoms. TM-align will make different mappings of equivalent atoms and then come up with the TM-Score for that mapping. It will look for the best TM-Score for all the mappings that it evaluates (it uses several heuristics for this). There are other methods as well such as CE to generate the mappings but TM-align and its cousin FR-TM-align are the most accurate automatic methods and are completely structurally based. The mappings are made to maximize TM-Score and not based on sequence identity.

If you already know what atoms to superpose from your sequence alignment, you can feed these to TM-Score which will give you your best answer. The main difference from RMSD is that TM-Score does not allow any divergent loop regions to dominate the superposition and measure of similarity. If you don't have those, it is not a big deal to use RMSD.

To what resolution have you solved your structures? The RMSD between your two isoforms may reflect imprecision in the model combined with genuine structural differences.

To objectively and quantitatively examine conformational similarity in solution without requiring superposition, I would suggest you consider small-angle x-ray scattering and the use of the structural comparison map see Hura GL et al., (2013) Comprehensive macromolecular conformations mapped by quantitative SAXS analyses. Nature Methods 10, 453-4. SAXS can be done by mail-in to the SIBYLS beamline from 96-well plates so is low cost and fast - see http://bl1231.als.lbl.gov

For comparing structural similarity you have lot of options:

For evaluation you can go through: Procheck which can clarify you about the stereo-chemistry of the both proteins and how stereochemically they are similar. In the same way, Errat Plot, Verify 3D, Prosa and RMSD also assist you to evaluate the structural similarities and differences of both the proteins.

Please follow the given link provided: In this manuscript you will find their use and how to interpret the results.

https://www.researchgate.net/publication/248382055_Cystathionine__Lyase_like_Protein_with_Pyridoxal_Binding_Domain_Characterized_in_Leishmania_Major_by_Comparative_Sequence_Analysis_and_Homology_Modelling

Article Cystathionine β Lyase like Protein with Pyridoxal Binding Do...

Interacting and Non-interacting protein pairs can classify the structural features.

You can use RAMPAGE and PorQ tools in the assessment of protein structures

You can refer to below attached paper full-text, which has all the methodology details and link to webservers.