PCCA+ eigenvalue convergence issue
Posted: Thu Oct 04, 2012 8:55 am
Hi everyone,
I'm new to msmbuilder, I've been trying to get a macro state model of barnase (1 microsec. simulation with GROMACS, saved every picosecond, in 10 xtc files each containing 0.1 microsecond trajectories) with various parameters. I've tried using backbone (I specified the atom indices) RMSD as my metric and built micro states with 2A, 2.5A, 3.5A and 4A clustering (hybrid), ended up with very reasonable implied timescale plots. I used the BuildMSM.py with a lag time of 1 as per the tutorial and tried both PCCA and PCCA+ algorithms to do the macrostate clustering and scipy ended up giving up on the calculation after a while with the error along the lines of;
(100001 iterations, 0/5 eigenvectors converged) [ARPACK error -14: DNAUPD did not find any eigenvalues to sufficient accuracy.]
for every RMSD separation. I should also mention that I've tried both MLE and Transpose algorithms on multiple different calculations (in which either MLE calculation was the same/very similar to the Transpose calculation, I've also had several occasions of Cluster.py dying without an error as well, mostly for another protein though).
After this I tried using a coarser set of trajectories and subsampled (using stride in the beginning, with the ConvertDataToHDF.py) my data. I tried every 10 step and every 100 step, both gave me much worse implied timescale plots and the result was the same (I also changed the metric and used the 'heavy' option of CreateAtomIndices.py, just to make sure that my indices are not the source of the problem). Lastly I tired this metric on the original set of data, without the subsampling, which also gave the same result.
Finally I should mention that I'm using EPD 7.3-2 as my python interpreter.
Can you give me any suggestions, I'm sure I'm missing something simple but I can't seem to figure out where I'm going wrong.
Best regards,
Ali Sinan Saglam
I'm new to msmbuilder, I've been trying to get a macro state model of barnase (1 microsec. simulation with GROMACS, saved every picosecond, in 10 xtc files each containing 0.1 microsecond trajectories) with various parameters. I've tried using backbone (I specified the atom indices) RMSD as my metric and built micro states with 2A, 2.5A, 3.5A and 4A clustering (hybrid), ended up with very reasonable implied timescale plots. I used the BuildMSM.py with a lag time of 1 as per the tutorial and tried both PCCA and PCCA+ algorithms to do the macrostate clustering and scipy ended up giving up on the calculation after a while with the error along the lines of;
(100001 iterations, 0/5 eigenvectors converged) [ARPACK error -14: DNAUPD did not find any eigenvalues to sufficient accuracy.]
for every RMSD separation. I should also mention that I've tried both MLE and Transpose algorithms on multiple different calculations (in which either MLE calculation was the same/very similar to the Transpose calculation, I've also had several occasions of Cluster.py dying without an error as well, mostly for another protein though).
After this I tried using a coarser set of trajectories and subsampled (using stride in the beginning, with the ConvertDataToHDF.py) my data. I tried every 10 step and every 100 step, both gave me much worse implied timescale plots and the result was the same (I also changed the metric and used the 'heavy' option of CreateAtomIndices.py, just to make sure that my indices are not the source of the problem). Lastly I tired this metric on the original set of data, without the subsampling, which also gave the same result.
Finally I should mention that I'm using EPD 7.3-2 as my python interpreter.
Can you give me any suggestions, I'm sure I'm missing something simple but I can't seem to figure out where I'm going wrong.
Best regards,
Ali Sinan Saglam