A short holiday break and a rather protracted bout of the flu have kept me from blogging. So I will link to an article of mine that just got published in the magazine of the Chemical Society of the Indian Institute of Technology (IIT), Delhi. The article is written for the layman and talks about the importance of realizing that flexible molecules have multiple conformations in solution. Such conformations cannot be determined by NMR alone due to their rapid interconversion.
In the article, I describe NAMFIS (NMR Analysis of Molecular Flexibility In Solution), a joint computational-NMR approach which can derive a Boltzmann population for flexible molecules in solution. This information can be very useful for deducing, for example, the protein-bound conformation of a drug. But it can also be useful under other circumstances where determining conformation is important, such as for organic molecules assembling on a surface. Comments, criticism and questions are of course always welcome.
Of course molecules at room temperature should be described by thermodynamic ensemble,not a single conformer. However, I don't like the tendency to overparametrize when people increase the complexety of the model. Surely, you can get much betet fit to experimental data by increasing the number of variables, but is it really science?
ReplyDeleteI'm about to head out for the night, but a few points -
ReplyDelete1.) What about extracting order parameters from NMR data to analyze dynamics at faster time scales than milliseconds?
2.) As someone who used to do solid state NMR for a living, molecules in the solid state are not static, especially if you're referring to small molecules/ligands bound within proteins. While the protein crystal/aggregate may be immobile on the timescale of the experiment, the small molecule bound does not have to be immobile.
I am pretty sure that I missed something in the article to make me mention the above, having only skimmed it. But anyway.
- MJ
That's a very good question. Over-fitting is certainly possible with too much data and two few conformations. In most of the cases we have looked at, we have had about 30-40 distances for complex molecules and 10-15 coupling constants which seems to be a reasonable data set to avoid over-fitting. Plus, the usual alternative to such a protocol, simulated annealing or molecular dynamics using NMR constraints, tries to essentially compress all the NMR data in a single conformation. That's over-fitting taken to its extreme I think.
ReplyDeleteAs for order parameters, that's an interesting question and I think one could probably use them. In principle we can use any kind of data including chemical shifts since boiled down to its essentials, the problems is simply a least-squares fitting
problem.
For solid-state NMR, I agree that the ligand may move about. How much it does so is an interesting question though.
Thanks for your comments! Interesting questions, all of these.
You may want to check out a paper from Ron Levy's group at Rutgers from last year on the conformational dynamics of a bound substrate in the active site pocket of cytochrome P450 at http://levygroup.rutgers.edu/pdf/JACS_129_474-475_2007.pdf - the computational study is based on order parameters determined by solid state NMR. Most of the "order parameters from solution NMR in biochemistry" material I'm remembering at the moment has to do with proteins (HIV protease and its flaps, for instance) and not small molecules, but that's probably just a function of the fact I'm a bit rusty when it comes to NMR.
ReplyDeleteTo an extent, the question of whether a ligand is mobile within an active site pocket/cavity may be more a question about the system under investigation than what method you use to probe it. For instance it should be noted that cytochromes P450 can have cavernous active sites that can bind multiple substrates - however, there are plenty of enzymes where that doesn't apply. If you think about membrane protein channels, in particular the ion channels, it's not unusual to find multiple binding sites to shuttle the ion/molecule through the channel. I think it's the potassium channel that has four binding sites in its selectivity filter, and the notion is that only two are occupied at the same time (site 1 & site 3, or site 2 and site 4).
More things to think about, at the very least.
- MJ
Nice article on NAMFIS (and very clear). It does seem like you're getting something for nothing. A little tweaking of a parameter here and there should get you just about anything. I think von Neumann said something like -- give me 5 parameters and I can fit an elephant on a curve, give me 6 and I can make it dance.
ReplyDeleteRetread
Thanks. Von Neumann's quote is priceless! It's hard to say whether your fit is meaningful or not. Usually the more the data the better, but only up to a certain point. In case of NMR data, people fit it to conformations all the time. One of the best ways to keep a check on overfitting is to see if your end model can predict data that was not included in building it. For example for peptides you may fit C (alpha) chemical shifts, but then you can double check by investigating whether you can predict carbonyl C or amide N chemical shifts.
ReplyDelete