I am working with small molecules that seem likely to facilitate pi-pi stacking interactions and/or halogen bonding. I am curious to know whether any small molecule docking/screening freeware takes these types of interactions into account. Furthermore, what are the best ways to characterize them in the resulting docking results?
Hi Mr Stewart
If i understand your mean, you can see both pi-pi interactions and halogen bonding in Autodock Vina result. Although you can set docking result by changing RMSD (Root Mean Square Distance ) .
Thank you. For clarification, I'm more curious if these interactions are taken into account by the scoring function of any freeware or server-based docking programs. Furthermore, how heavily do they weigh these types of interactions? It is my understanding that both Autodock tools and Pymol are suited for analysis of docking outcomes for autodock vina. UCSF Chimera also has some utility for Swissdcok but I'm less familiar with this program.
dear john
your mean its not clear, if you wanna help, you can explain more .
Force field based calculations do not evaluate electron dependent properties and, for that, you will need quantum mechanic calculation. It is possible that, for some force field, specific terms have been included trying to describe these interactions, but I am unaware of that. At least for the most common force fields dealing with proteins, that is, GROMOS, CHARMM, AMBER and OPLS, in its most common versions, there is no specific term to deal with the interactions you mentioned.
John,
to the best of my knowledge no server based docking app has a term in its scoring function that attempts to mimic halogen bonding. The more relevant question is, what is the problem you are trying to solve that might be affected by including or ignoring halogen bonding? Are you interested in pose prediction, virtual screening, relative affinity estimation?
Actually pose prediction, virtual screening, and relative affinity are all all facets of what we want to do. We have a set of mutational variations which have highly conserved aromatic residues. We know these residues must interact with some aryl halides that can inhibit all mutational variants. However, the programs we have been using (Autodock vina and SwissDock) don't seem to take pi-pi interactions or halogen bonding into account; or rather they consider them repulsive rather than attractive interactions. Although there are programs that can take these interactions into account, at this point we would like to use something that is highly accessible.
There is no good evidence that including halogen bonding terms will improve VS performance by docking tools. For the other two problems however there may well be an impact that might even be positive ;-). If you have good system specific evidence that these terms matter in your problems, then you may have to look elsewhere for your scoring function.
just because aryl halides make good inhibitors does not mean halogen bonding is the cause; the aryl halide inhibitors of serine protease work probably work due to water displacement in a highly geometrically specific fashion, not because of halogen bonding.
An important point on virtual screening studies is the low capability to predict affinity. The orientation is well predicted by the most used programs, but affinity is a completely different question (and much more complex). On specific receptors, and specific series of compounds, the prediction may be high, but looking as an average between different receptors and a diverse set of compounds, the correlation between experimental and theoretical affinity is not much larger than 50% - you will find several reviews on that matter, as it is a hot topic in literature. So if you are planing to study a small set of compounds, be aware of that. Of course, if you are dealing with thousands of compounds, your chance of finding something interesting increase.
One additional point: even not having specific terms to describe such interactions, it is not unusual that other terms compensate this absence due to adjustment to empirical properties. For example, some works in literature demonstrated that the molecular mechanics derived (using the most established force fields for protein modeling) interaction energy on a CH-pi interaction between aromatic rings and carbohydrates presents a linear correlation to the quantum mechanical derived.
So, always as possible, build a training set of compounds (as similar to your compounds and as large as possible), with known affinities, in order to offer an objective description of how your experiment is producing the expected results. The quality of the predictions is too system dependent.
I totally agree with Hugo,
in Virtual Screening is possible to achieve some level of enrichment but in practice there is no correlation between docking score and experimental affinity (and this is true for docking in general, not just VS).
In this respect I think that the following publication clearly investigated the issue.
http://www.ncbi.nlm.nih.gov/pubmed/17004707
- Badges
- Science topic