Hello I'm trying to build 3D model for ALS2 (ALSIN) protein. The websites like SWISS-MODEL can't give proper model. Also, I can't use I-TASSER because the protein is longer than 1500 amino acids. Which software do you suggest to draw proper 3D model for this protein?
We keep going over this - on seemingly the same topics and seemingly with the same posters - and nothing changes.
The websites like SWISS-MODEL can't give proper model.
What does this mean exactly? It is difficult to provide advice before you tell us why SWISS-MODEL did not work for you. If this is a continuation of your previous question where you were trying to model something without a template, the answer is the same: modeling a protein - or part of it - without a homologous template can't be done reliably in most cases. This goes for any program, not just SWISS-MODEL. Modeling a large protein without a template is almost impossible to do reliably. Again, this goes for any program. To be clear, >1500 amino acids is considered large, which is why I-TASSER won't even humor us by attempting to model.
Two options that I would recommend, in order of my preference: 1) model only parts of the protein for which you have homologous templates, and forget about the rest; 2) find out about a domain organization of your protein, divide it accordingly, and submit individual domains (presumably smaller than 1500) to I-TASSER. This option will give you some result for everything, but don't expect magic for regions without homologous templates (which brings us back to my suggested option #1).
If you are trying to model a protein without an available template you could try our method DMPfold, which is available as a web server here and described in this paper.
Template-free modelling has improved considerably in the last few years, but just submitting a 1500+ residue protein is unlikely to give a decent model. Consider splitting it up by predicted domains, or just modelling the part you are interested in.
I think this is a good general advice, and @jgreener would know better than most of us here since he is in the lab that has been among the pioneers of covariance-based modeling. Still, the first goal of covariance/distance-based modeling is to get the fold right, and the model will generally be considered accurate if it is within 2-3A RMSD of the real structure. That is good enough for general fold assignment, molecular replacement applications, cryo-EM density fitting and maybe even coarse docking studies. Not sure that it would be good enough for studying the effect of point mutations on folding, which is what the original poster was asking about. It certainly doesn't help that the protein in question is large and multi-domain in nature.
Yes, I would broadly agree, getting the fold right is hard enough let alone predicting the effects of variants. There has been surprisingly little work in assessing the utility of de novo models for variant effect prediction, but it is something we are working on.
I wouldn't expect individual structure prediction runs with a single mutation to turn up much difference, but de novo models can still be useful for predicting and interpreting variant effects. For example, the de novo model will show whether a site to mutate is buried or exposed more accurately than prediction of accessible solvent area. It will also give an idea of how much space there is to fit a mutated amino acid into. See some work by Mike Sternberg's group on doing this with template models here - we are now reaching a level of accuracy with de novo models where this is becoming possible.
Thanks a lot @jgreener. I tried your website which only accepts the 500 residues in each run. I checked the functional domains of ALSIN protein in the NCBI's Conserved Domain Database (CDD). The result is below [https://www.ncbi.nlm.nih.gov/Structure/sparcle/archview.html?archid=13421284][1] As you can see there are multiple domains and don't cover the ALSIN protein sequence completely .So is it possible for me to obtain the sequence of each domain from CDD results and insert them desperately to your web server and then merge those models? There are two main problems in this. First, they don't completely cover all ALSIN protein sequence. Second, how may i be able to merge these models together? I know it won't result in a proper model but i need it as a demonstration for my article. Also, I've found an article which has a model from ALSIN protein but it is unclear how they have obtained this model. I provide the link to that article below. [https://www.ncbi.nlm.nih.gov/pubmed/17955197][2] I will be so pleased if you could guide me to solve this problem.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
Thanks a lot for your complete response. Yes, I exactly meant the problem that you mentioned in my previous question which was about another protein. The SWIISS-MODEL can not give good model because of homology modeling problem. I want to analyze the effect of a mutation on this protein. The I-TASSER suggested the same, to insert the domains of this protein separately, but in this case is it possible to merge these separate domains together and build a single model? Or I need to report the domain simulations and effect of the mutation of its folding separately?
I don't mean to sound like a broken record, but this can't be done reliably. It can be done for illustration, but not for doing serious science. Think of it as solving any complex problem: if we can't solve individual sub-problems well, how likely is it that putting those together will result in a good solution?
Back to my response from our previous interaction: if you can't build a good homology model, adding a mutation to it will not give you reliable answer regarding folding. If you have a good homology model, adding a point mutation to it may still not tell you much in terms of stereochemistry. Studying protein energies may be the only way to do it, and it still would not be straightforward.
In case I have not been clear up to this point: what you are trying to do is very unlikely to succeed. Let's say you had a good crystal structure of your whole protein - so no need to build any model - and were only studying the effects of point mutations. That would be difficult enough on its own. Building a homology model of your protein - assuming the template was available - and studying mutations from that model would be even more challenging and less likely to work. But you are not even close to that category. Though you are not telling us details as to what parts of your protein, if any, have homologous templates, it seems that a good chunk of it has to be modeled ab initio. That would be another level of difficulty and is further compounded by your stated objective to study how mutations affect the folding. I don't see how a professional protein modeler would succeed here with a protein that can't be modeled reliably in the first place. To reiterate: none of this has to do with you, or with programs you are using.
I really appreciate your kind and useful response. It completely changed my thought about the protein modelling. Also i want to add that my main purpose is only a simple illustration of model. because my mutation (IN ALSIN case) has resulted in a stop codon with truncated protein. But in other Missense mutations i will consider your instruction for sure.