Dear Marisol,

You got tons of good advice above so use it well.

However, in all this rush to help you people did not mention two preconditions.

1) You should try to ascertain first, whether your protein belongs to a well foldable class. A modern view of proteins tells us that the protein kingdom extends from very well folded (stable three dimensional structures) to completely unfolded all the time (unstable).

2) Which parts of your protein can be predicted by whatever method is available with what accuracy. That includes checking for presence of disordered elements and the domain structure.

RES 1) A quick shortcut to the first question is asking whether your protein/sequence is crystallizable. There are several web services on line. I like XtalPred. It gives you a lot of auxiliary information like secondary structure predictions, or transmembrane or disordered elements. The most significant it classifies your sequence into 5 classes. It tells you whether it is easy or very difficult to do it. This is a proxy to the statement whether it is uniquely folded. An additional advantage is that it produces good homologs in all crystallizability classes. If you do not find your protein there you might be out of lack.

RES 2) You should run Blast searches for homologs but also run FFAS03 to find remote homologs or any other site (like Nik Grishin's). Run secondary structure predictions and some domain structure predictions. Maybe you will have more luck with a single domain predictions. A lot of proteins have significant mobility imparted in them (this is the way they operate) therefore there might be lack of appropriate templates.

The main result of the structural genomics centers is that there is a significantly slower rate of discovery of new folds so there is a substantial chance that your protein or some of its domains are already known and the only challenge is to recognize it.

Good Luck

Rosetta will be the best at this moment. http://robetta.bakerlab.org/

i know not that type the predictio, but searching the crystal of a similar protein you can predict it. Regards ... if everyone knows that program I'm interested too :)

If you are mentioning about homology modeling method, there are few programs that allow you to predict the 3D structure of a protein given the amino-acid sequence. You can check SWISS-MODEL, MODELLER. You can also have a look the server HHpred where they integrate sequence database search with MODELLER. This server is one of the highest rank servers in CASP competition.

Hope it helps!

you can use different methods: Homology modeling (HHpred), Protein threading (I-TASSER,) and Ab initio protein modelling (I-TASSER, robetta), so you have to choose which methods is useful to you

Thanks so much to all of you for your feedback. I will check the programs and let you know which one was the best!

To guess the 3D structure without likewise known crystal is a very interesting topic.

You might be interested in QUARK, it is the best template-free program for protein structure prediction.

It has the disadvantage of allowing only proteins of billions).

I'd use Quark, it uses a hybrid strategy, first an heuristic approach for finding the most-likely-to-occur conformation for small fragments of your protein, and then it assembles the structure using Replica Exchange Monte Carlo searches for the final structure. No crystal needed.

http://zhanglab.ccmb.med.umich.edu/QUARK/

Did you checked whether your protein has similarity to any protein structures available at PDB. You can BLAST your protein against PDB database...

Yes Boopathi i have tried, but not good results

Thanks victor i will definitely try that program

Thanks to all again :)

Marisol Romero, it is possible to do homology modelling of the protein with less then 20% homology to the template. I did that myself it worked good.

The Dunbrack program MOLIDE may do what you want. I do not recall if it starts by Blasting the PDB with the query sequence or if you need to do that manually.

I suggest you also look at new results from Chris Sander's lab which look quite good if you have homolgous sequences.. See these recent papers:

Three-dimensional structures of membrane proteins from genomic sequencing.

Hopf TA, Colwell LJ, Sheridan R, Rost B, Sander C, Marks DS.

Cell. 2012 Jun 22;149(7):1607-21.

Protein 3D structure computed from evolutionary sequence variation.

Marks DS, Colwell LJ, Sheridan R, Hopf TA, Pagnani A, Zecchina R, Sander C.

PLoS One. 2011;6(12):e28766.

Hi Marisol,

Did you try SWISS-MODEL? I think nobody mentioned it so far, just did not want repeat anyone's suggestion. SWISS-MODEL would require a homology model, i.e. known structure and your sequence. The method is not really new but might be helpful. Are you trying to crystallize the protein?

Good luck,

Andrei

Prediction can be homology-related, using other closely related proteins as templates. If you find a closely related protein, you can do the residue mutations on SWISS-MODEL.

Quick question, would this be a transmembrane protein by any chance?

Try phyre2. My protein didn't match to anything but I still something out of it. Of course the accuracy might not be good but at least you have an idea.

Dear Marisol,

you can visit the Protein structure predection center website ( http://predictioncenter.org/ ) to find out which methods/servers are useful and provide accurate predictions under which limitations. They have organised a accesment of methods named CASP ( Critical Assessment of protein Structure Prediction ) which evaluates comparative and de-novo prediction methods.

The results of CASP 9 are published in in a scientific journal Proteins: Structure, Function and Bioinformatics Volume 79, Issue S10, Pages 1-207 (2011).

Here you find the best methods currently available also for hard targets with low sequence identity to known structures.

Happy modeling,

Marco

Identity less than 20%. Used I-TASSER and HH-pred and it worked. Here's my

pubblication: http://dx.doi.org/10.1016/j.jmgm.2011.04.008 ... I hope it'll be usefull.

Rafael, the context of the title suggests alternative experimental methods, but the body of the post wishes to use amino acid sequence for 3D structure prediction.

Marisol: Is this for all in-silico, or are you using in silico to guide experiments?

What about to measure a CD spectrum of your protein...? takes only some minutes, needs only some few µg of your protein and gives you the global secondary structure of your protein.

You can search your protein in structure data base, e.g. swiss model PDB server. If any similar stucture is available then you can do modelling.If your protein is completely new, protein structure can be determined by Crystallization, however, If protein is small in size (

One word answer "Rosetta"

Anyone else a fan of the M4T server?

http://manaslu.aecom.yu.edu/M4T/

The most popular way is homology modeling, which search for a 3D structure of the homolog of your target protein to predict the structure of yours. I was using the program MOE to do this, it is quite convenient.

Let me answer your question from a more theoretical point of view. (You might know this by now, if you have searched for the programs that have been suggested.) The idea is the following. A real protein will fold due to the electrical forces in the molecule. If one can compute a state that is a local or global energetic minimum of the molecule, this is likely to be the state the real molecule is in after it has folded

(itself). In this state, all forces will be balanced. There are programs which attempt to find such a state.

Several of these have been suggested above.

u cn use MODELLER for it. But disadvantage is that it is not free..u also use PS2 server which is free

Jinwon Jung yesterday I tried Robetta, but it is showing that it will take 440 days to complete the QUEUE. Does this usually take that long?? I cannot wait that long. Thanks

Have you tried to align your sequence against a proteins database? It should give you important clues. BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi) is a good server I used a lot for similar purposes. Then, I found Dirk Windisch's suggestion absolutely appropriate. If you can, try to collect a CD spectrum to exclude/confirm some of the matches you have found with BLAST. Finally, when you have found "the best" analogue of your protein, you can use MODELLER (http://salilab.org/modeller/) to built a 3D structure for your protein on the basis of that of the analogue.

Thank you all for your suggestions and comments! I am using I-Tasser (for more than 200 residues) and it works really good. I would recommend it to all those who want to have an idea of how your protein would look like in reality. Best!! =D

Hi,

You may like to have a look at these two papers

You may try to predict the 3D structure using a number of tools, I suggest you read:

http://www.cicy.mx/Sitios/Journal/pdf/2011/October/HighDefinition/IBCJ_2011_1_1_26S.pdf

http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0012483

They are open access

Obtaining the models may be the easy part, the hard part is to rate its quality and appropriateness to determine if it can be used to predict some protein's properties.

Best wishes,

Rogelio

The best protein structure prediction method is homology modelling. The way is already been described by Mariano and MODELLER is best at Homology Modelling. Another way is the protein fold recognition method, PHYRE server is good at it. Third method is ab intio structure prediction, which is of I-TASSER and QUARK. I have used the I-TASSER 1.1, it's awesome.

Homology Modelling is the best and accurate way of PSP as one can analyze the function by taking into consideration of template.

You can check with Pepstr, a new tool to predict the 3D structure of Small peptides

http://repository.ias.ac.in/43075/1/28-PUB.pdf

You can choose various method, depend on your sequence homology to the protein available in the experimentally 3D structure.

1. > 40% homology- homology modeling is the best , use Modeller

2. < 40% homology- threading method- Mgenthreader or Itasser

- Ab-initio method - rosetta/robetta - if your sequence

without having the crystal, you can have the 3D structure of protein/pepide by NMR spectroscopy, which is generally done in solution phase. You can check that in PDB also so many 3D structure reported on the basis of NMR Spectroscopy.

Hi! Try hhpred http://toolkit.tuebingen.mpg.de/hhpred. It gives you a blast and makes a model with the best template automatically. It is not much accurate but gives hou some place to start.

Hi, you have to choose a program or method (homology, threading or AB-initio) according your protein sequence. in this link you can see different links and/or programs http://en.wikipedia.org/wiki/List_of_protein_structure_prediction_software.

I know, is wikipedia but the articlue is useful.

Hi,

Does someone know what I-TASSER stands for? the meaning of the letters, or is just a random name?

Thanks

Maybe, this approach will help you:

Correlation between sequence, structure and function for trisporoid processing proteins in the model zygomycete Mucor mucedo

Author(s): Ellenberger, S (Ellenberger, Sabrina)[ 1 ] ; Schuster, S (Schuster, Stefan)[ 2 ] ; Wostemeyer, J (Woestemeyer, Johannes)[ 1 ]

Source: JOURNAL OF THEORETICAL BIOLOGY Volume: 320 Pages: 66-75 DOI: 10.1016/j.jtbi.2012.12.009 Published: MAR 7 2013

Abstract: Terpenoids, steroids, carotenoids, phytoenes and other chemically related substance groups fulfill multiple functions in all realms of the organismic world. This analysis focuses on trisporoids that operate as pheromones in the phylogenetically ancient fungal group of mucoralean zygomycetes. Trisporoids serve as pheromones for recognizing complementary mating partners and for inducing the differentiation program towards sexual spore formation. Trisporoids are synthesized by oxidative degradation of beta-carotene. Structurally, they are related to retinoids in mammals and abscisic acid in vascular plants. In order to evaluate evolutionary relationships between proteins involved in trisporoid binding and also for checking possibilities to recognize functionally related proteins by sequence and structure comparisons, we compared representative proteins of different origins. Towards this goal, we calculated three-dimensional structures for 4-dihydromethyltrisporate dehydrogenase (TSP1) and 4-dihydrotrisporin dehydrogenase (TSP2), the two proteins involved in trisporic acid synthesis that have unequivocally been correlated with their catalytic function for the model zygomycete Mucor mucedo. TSP1 is an aldo-keto reductase with a TIM-barrel structure, TSP2 belongs to short-chain dehydrogenases, characterized by a Rossmann fold. Evidently, functional conservation, even implying very similar substrates and identical cosubstrates of enzymes in a single organism, turns out to be essentially independent of basic protein structure. The binding sites for NADP and trisporoid ligands in the proteins were determined by docking studies, revealing those regions affecting substrate specificity. Despite the pronounced differences in amino acid sequence and tertiary structure, the surfaces around the active sites are comparable between TSP1 and TSP2. Two binding regions were identified, one sterically open and a second closed one. In contrast to TSP1, all docking models for TSP2 place the trisporoid into the second, channel-like region. (c) 2012 Elsevier Ltd. All rights reserved.

Marisol. ... maybe this help you http://www.jove.com/video/3259/a-protocol-for-computer-based-protein-structure-function

Dear Marisol,

You got tons of good advice above so use it well.

However, in all this rush to help you people did not mention two preconditions.

1) You should try to ascertain first, whether your protein belongs to a well foldable class. A modern view of proteins tells us that the protein kingdom extends from very well folded (stable three dimensional structures) to completely unfolded all the time (unstable).

2) Which parts of your protein can be predicted by whatever method is available with what accuracy. That includes checking for presence of disordered elements and the domain structure.

RES 1) A quick shortcut to the first question is asking whether your protein/sequence is crystallizable. There are several web services on line. I like XtalPred. It gives you a lot of auxiliary information like secondary structure predictions, or transmembrane or disordered elements. The most significant it classifies your sequence into 5 classes. It tells you whether it is easy or very difficult to do it. This is a proxy to the statement whether it is uniquely folded. An additional advantage is that it produces good homologs in all crystallizability classes. If you do not find your protein there you might be out of lack.

RES 2) You should run Blast searches for homologs but also run FFAS03 to find remote homologs or any other site (like Nik Grishin's). Run secondary structure predictions and some domain structure predictions. Maybe you will have more luck with a single domain predictions. A lot of proteins have significant mobility imparted in them (this is the way they operate) therefore there might be lack of appropriate templates.

The main result of the structural genomics centers is that there is a significantly slower rate of discovery of new folds so there is a substantial chance that your protein or some of its domains are already known and the only challenge is to recognize it.

Good Luck

Hi Marisol I-TASER means I-TASSER: iterative threading assembly refinement algorithm.

Metaserver is a good resource that returns prediction results from multiple servers and you just submit it to one central portal.

http://www.ncbi.nlm.nih.gov/pubmed/11524381

https://genesilico.pl/meta2/

Here's a complete list of structure prediction software servers

http://en.wikipedia.org/wiki/List_of_protein_structure_prediction_software

I agree, Rosetta seems to be the best but you can use also Modeller if you have homologous.

Dear All,

My experiences with Rosetta/Robetta are not as exhilarating as opinions expressed here. A mixture of protein attributes as described by me in other threads (partial solid/ partial liquid) makes exact the predictions almost impossible. So the question remains what are people experiences (beyond the hype). What was the best result people were able to get as measured by the fold and RMSD versus the percentage of similarity and known three-dimensional template. I am interested in how many people got reasonable answers with ab initio modeling (no identifiable template).

First, do protein-protein BLAST and check for homologous structure similarity if you get homologous structure then used that structure as a template and perform modelling using MODELLER but it only possible when sequence similarity is 30-35% otherwise you have to go for loop modelling "FALC loop" server or ab-initio based I-tasser, Quark .

Dear Nivedita,

Your answer is the best example what is the difference between the hype and the reality. Existence of solid and liquid like portions translates to the possibilities that folds can be different even when we have high sequence identity and they can b very similar when they are quite remote. Modeller is just horrendous if it comes to insertions and deletions. It is doing relatively well in well conserved regions. Rossetta/Robetta has such a strong bias towards the know structural fragments that will always bias towards what was seen not what prevails in the conformational ensemble. In my hands I never obtained a model that could have been used successfully in molecular replacement even when the sequence homology was high. Molecular replacement is quite demanding but.. still... I like I-tasser as it originated from a long history of discrete modeling.

I agreed with Boguslaw I-TASSER is better then Modeller and Robbeta because for the first time, the sequence-based contact predictions from machine learning techniques are found helpful for both template-based modeling (TBM) and template-free modeling (FM) by I-tasser and also iterative template fragment assembly simulations function inslights are derived by matching the 3D models

I-TASSER if your sequence length is more that 200 aa or you can go for QUARK. Besides intensive modelling of PHYRE2 is also recommended to obtain a good structure.

Dear Marisol,

certainly you've had a flood of suggestions. The real problem is not this or that program to use. All programs are good and all programs are bad. Only experience will tell you which program works best in a certain condition. What is important is the set of steps that must be addressed in order to have a correct approach to modeling. I suggest the following recipe, Mediterranean style and rather simple.

Recipe for modeling

1) the sequence is that of a protein known to be globular? if the answer is yes proceed, otherwise you hardly will have an acceptable model (you must check the quality of the sequence. The presence of charged stretchs, small, polar and hydrophilic amino acids makes difficult any modeling)

2) Once you have the models, choose the best one and submit it to molecular dynamics in the presence of water (water box) until equilibrium is reached (and also something more). If the structure of your model is still compact go to checks.

3) The model thus obtained should be tested not only for its energetic quality (proCHECK, etc..) but also for its globularity. The globularity is not a property, but a number of empirical characteristics that all globular proteins must have (there are several works around). You might have a model energetically valid but not globular.

After all this, maybe you have a reasonable model. Good luck.

A series of good answer is given which will definitely help you for further analyzing your results and predicting your 3D structure in the absence of crystal structure. I would like to summarize few vital points here which may help you in deciding your approach in this context.

3 approaches are broadly available.

1. If there is a better template of close identity (more than 35% on an average, higher the better), similar length and good resolution the better way is to go for homology modeling.

2. Your interest lies in particular region of protein and you want to predict the structure only for domains or particular folds, the template you are getting not fulfilling the criteria provided in point 1, you can use fold recognition methods using any online server.

3. Neither point 1 or 2 is properly getting satisfied and you are almost clueless about the structure of the protein, ab initio is the most preferable choice in this aspect.

I had used I-TASSER for ab-initio prediction method. But you can try first SWISS-MODEL server and 3DPSSM before going to ab-initio method.

Dear Marisol, I am not sure if you have already succeed in your plans to get a protein model, but as mentioned above there are several methods by which you can obtained a structure. If you still need some help or advise you can always contact me: [email protected].