Dear all,
My study is using Molecular Dynamics Simulation to investigate biological or chemical systems. My recent study is testing the stability of BSA at different conditions. I find Bioinformatics is focusing on the analysis of protein structures. And I find DSSP is a good way to analyze the secondary structure change with time. But I need a method to compare BSA in two different conditions. The dynamics fluctuations of DSSP are not suitable for this task. I need a method to generate the average or overall conformation of the protein during a time period and output the regions that have difference.
I uses AMBERTOOLS to generate the average structure of the protein during the simulation, and then use one as the reference to compare with the other one. The result was not clear to show the difference.
Could you give me some suggestion that I could use to compare BSA at two different conditions? You may give me some names of methods, and I will look into them by myself.
Thanks,
Chris.
If you expect (not necessarily easily discernable) changes in global conformation of the protein, then trajectory projections onto principal components' subspaces (derived from PCA or tICA) are often the way to go for the visualization of conformational transitions.
Look up e.g. the link for reference.
http://scitation.aip.org/content/aip/journal/jcp/141/1/10.1063/1.4885338
Thanks, I am reading the paper.
And I find the website http://thegrantlab.org/bio3d/webapps supplies various analyses for protein structures.
One among the standard methods is the multicomponent analysis such as the Principal Component Analysis (PCA). Note that the use of the natural internal coordinates (standard chemical geometrical parameters ) makes it easier to be understood. Ramachadran Phi and Psi angles are also useful.
Thank you, Dr. Alix.
I am using AMBERTOOLS PYTRAJ to calculate PCA of the BSA protein during a simulation. It's interesting that I get the figure but don't understand what it means.
I am reading the manual to figure out how PC1 and PC2 are calculated.
I´m not a specialist in PCA, but well involved with this method at the moment.
From this plot it is very hard to tell what happens exactly. Normally you have to analyze the individual primary components to detect the exact parameters that change. The PCi components are weighted with a downward percentage with increasing i, and component PC1 therefore gives the largest changes with the highest percentage of relevancy. Maybe you have to deconvolute your analysis to the individual components to understand it better. Principally PC1, PC2, ..., PCn would also give different aspects of your structural change of BSA.
Often, you can visualize the individual components to see whether they can be directly interpreted as bending/opening-closing motions. That is, you take the average structure and add/subtract the PC in question to the respective coordinates to directly visualize the motion.
Also, what you could do is calculate your PCs based on all trajectories combined, and then inspect and compare 2D projections (as the one you attached) of individual trajectories corresponding to a given set of conditions. This way, you might find out that one set of conditions favors e.g. large values and another favors small values along PC2, and then interpret this based on prior analysis of PCs (see above).
Note that if your simulations are too short, you might not get adequate sampling and your PCs will mostly capture meaningless fluctuations. So getting enough data for analysis is a prerequisite.
If you know a bit of Python, you can follow and modify the workflow suggested for clustering with MSMBuilder (see link). They allow you to either use PCA or a variant of PCA called tICA; the difference is that while PCA captures modes characterized by largest variance (largest correlated motions), tICA captures modes characterized by slowest decorrelation (slowest correlated motions). These two often give similar results, but on occasions (e.g., wildly fluctuating protein termini that would dominate the PCs) tICA will be more robust.
http://msmbuilder.org/3.5.0/examples/Intro/Intro.html
Thank you all very much. I am working on this analysis and learning MSMBuilder.
Have a nice weekend!
Experimentally: CD (sensitive but not very information-rich), 2D NMR (information-rich and sensitive), X-ray crystallography (laborious but can be very information-rich), electron crystallography (getting better all the time).
Computationally: the tools already mentioned.
- Badges
- Science topic