PCA projection crystal structures and MD
Issue #478
resolved
Hi, I am trying to project the first few PCs from protein X-ray structures over MD simulation data. When I perform a separate analysis, two different regions appear to contribute to the PC analysis, I mean, if PC1 & PC2 of crystal structures receive a maximum contribution from loop 1, PC1 & PC2 from MD simulation seems to receive from another loop. I am not convinced that usual projection of the first PCs is helpful in that case. Is there is a better way to tackle this problem. Thanks
Comments (5)
-
-
reporter
Thanks, Xin, According to the DHFR tutorial part one on NMA, we first need to do the PCA on the structures and then on the MD trajectory and project the x-ray structures to the trajectory. My understanding is that what you project is the PC space of one of them over the other after using a common structure as a reference frame for fitting. Now, if the first 2 PCs from the PCA on the crystal structures describes a distinct type of motion (I mean different regions contributes to the first PCs of the PCA on structures and the MD simulation), can I still project the first few PCs even though what they describe is different. Simply, I need a way to project the structures on the MD trajectory forgetting about the PCs of the structures.
Any suggestion, Regards, Marawan
-
For "projection", I mean a linear transformation of one set of coordinates (such as Cartesion) into the other (such as represented by PCs). You don't need to do two sets of PCA. Just do a PCA on e.g. X-ray structures and use the function
project.pca()to convert the MD trajectory into a set of coordinates represented by the X-ray PCs. Then, in the plot, use the "z" component of the x-ray pca results and the converted MD coordinates, and you can see how they overlap (as illustrated by Fig. 8 in the DHFR tutorial). Key things you should pay attention are:- You need a proper alignment between x-ray structures and the sequence for MD, in which no gap exists in MD sequence at the positions where the PCA is performed.
- All MD frames are fitted to the same reference structure that was used for PCA before calling
project.pca().
To compare the essential dynamics captured by X-ray and MD, you do need to do PCA twice. In this case, you compare PCs using the
rmsip()function.Please check the DHFR tutorial NMA part I, 'Compare with MD simulation', for more details.
-
reporter
Got it, many thanks Xin, really appreciated.
Regards, Marawan
-
- changed status to resolved
- Log in to comment
Hi Marawan,
I am not sure I understand your question. Do you mean you have done PCA separately for X-ray and MD and try to put them together for comparison? If so, it is incorrect. You have to do PCA on X-ray structures alone and project your MD into the PC subspace, or vice verse.
It is still okay to do the projection if MD reveals some motions not captured by X-ray PCA, as long as you understand what you are doing and don't over-interpret it. For example, if you think the top PCs of X-ray PCA are better "reaction coordinates" (e.g. representing functional dynamics), you can use them to characterize the "state" of MD by projection, although MD may mainly explore other dimensions orthogonal to PCs (which are probably not function relevant).
Hope it helps...