I'm trying to figure out Protein-Protein interaction of two proteins without known crystallography structures. Any way software can predict this without the known structure? The problem is most software that does it is just textmining through articles, I was seeing if they can do some sort of prediction based off of just known sequences and other trivial things?

  • $\begingroup$ What options have you tried so far? There a lots of options, but it depends on what biological question you're asking, and the resources available to you. $\endgroup$ – James Jul 31 '16 at 10:49
  • $\begingroup$ Based on what I know you won't be able to run docking simulations on a sequence. The reason being our technologies for predicting 3D structures from sequences are not accurate enough to determine biological activity. Currently there is a competition every year for protein structure prediction software and I'm almost certain if someone comes up with a 100% accurate one there will be a nobel prize in it for them. Software that could predict structure from sequence accurately would radically accelerate molecular biology research. $\endgroup$ – Slawomir Andrzej Dziemborowicz Jul 31 '16 at 19:31

This review about protein docking will hopefully get you started.

Xue, L. C., Dobbs, D., Bonvin, A. M., & Honavar, V. (2015). Computational prediction of protein interfaces: A review of data driven methods. FEBS letters, 589(23), 3516-3526.

There is a page where they provide a short description about the different algorithms used, as well as links to the web servers. I also believe that most of these website should have a downloadable source code that you could compile and use on your own computer/cloud.

Here is a brief summary of the main methods highlighted by the paper:

Homology-based methods. In this approach, biological properties of a query protein are inferred from its homologs. The assumption here is that homologs share significant similarity in their sequence, structure and functional sites. Therefore, this approach performs best when close homologs are available.

Template-free machine learning methods. In this approach, machine learning algorithms are used to classify a target residue as either interfacial or non-interfacial, based on properties of both the target residue and the neighbouring residues. This approach is best used if there is limited availability of homologs, or when templates are not available or of poor quality.

Conclusion: If you have the sequence data available, then the 2nd approach seems most appropriate. You could also apply both methods and see the extent of agreement/disagreement between the methods, i.e, you could use the machine learning approach to detect the interfaces, and subsequently validate your docking predictions based on sequence data, if close-enough homologs are available.

  • $\begingroup$ Great succinct answer for a pretty vague question! $\endgroup$ – James Aug 4 '16 at 4:20

Not the answer you're looking for? Browse other questions tagged or ask your own question.