Closed marscher closed 8 years ago
marscher
commented
8 years ago For all lagtimes the off rates can not be computed:
koff_stats = []
from scipy.linalg import LinAlgError
for M in its.models:
try:
koff_stats.append(unbinding_rate(M))
except LinAlgError:
koff_stats.append(None)
print koff_stats
[None, None, (inf, inf, inf), (inf, inf, inf), (inf, inf, inf), None, None]
franknoe
commented
8 years ago T is clearly not fully connected. State 4 is absorbing.
I guess that means that this subset of data is actually disconnected. The new HMM code provides much better convergence to the optimal solution, so it reveals a problem that hasn't been seen before. I think we have a larger set of trajectories (at least 20 microseconds), we could try with those.
Since this is an issue with the data in the notebook rather than with the code, I would suggest to disable this computation of dissociation times and binding affinities for the purpose of this micro-release (it just affects the docs). We can find a more satisfactory solution for 2.2.
Please let me know when the notebooks output has been updated, I'm happy to check them.
Am 27/04/16 um 17:03 schrieb Martin K. Scherer:
During testing the trypsin_benzamidine_hmm notebook I've encounter the following problem. For lag time=1, the evaluation of kinetic/thermodynamic quantities the mfpt method fails for matrix A being singular. However T is fully connected and reversible.
@nsplattner https://github.com/nsplattner suggested that the new bhmm version gives a converged estimate, which is in this case not converged because we only feed in a very small data set.
We could handle the exception in the notebook, but I'd prefer a handler in the MSM.mfpt function (just to give an informative message, when we know how this is caused).
In the notebook cell 22:
dG_stats = np.array([binding_free_energy(M) for M in its.models])
kon_stats = np.array([binding_rate(M) for M in its.models])
koff_stats = np.array([unbinding_rate(M) for M in its.models])
print M
for M in its.models: print M.is_reversible print M.transition_matrix unbinding_rate(M)
output:
True [[9.90278363e-01 2.35753316e-37 4.67501507e-03 0.00000000e+00 5.04662171e-03] [3.19593056e-37 9.93143278e-01 6.85672226e-03 0.00000000e+00 3.81226426e-29] [1.17518596e-03 1.27145389e-03 9.93183749e-01 3.43529433e-04 4.02608166e-03] [0.00000000e+00 0.00000000e+00 0.00000000e+00 1.00000000e+00 0.00000000e+00] [3.80537587e-03 2.12050979e-29 1.20769094e-02 8.94882804e-04 9.83222832e-01]] bound state3 unbound state2 mfpt:A [[9.72163678e-03 -2.35753316e-37 -4.67501507e-03 0.00000000e+00 -5.04662171e-03] [-3.19593056e-37 6.85672226e-03 -6.85672226e-03 0.00000000e+00 -3.81226426e-29] [0.00000000e+00 0.00000000e+00 1.00000000e+00 0.00000000e+00 0.00000000e+00] [0.00000000e+00 0.00000000e+00 0.00000000e+00 0.00000000e+00 0.00000000e+00] [-3.80537587e-03 -2.12050979e-29 -1.20769094e-02 -8.94882804e-04 1.67771681e-02]]
LinAlgError Traceback (most recent call last)
in () 6 print M.is_reversible 7 print M.transition_matrix ----> 8 unbinding_rate(M) in unbinding_rate(bhmm, conf) 49 i_unbound, i_bound= index_unbound_bound(bhmm) 50 # MLE ---> 51 mfpt_mle= bhmm.mfpt(i_bound, i_unbound) 52 koff= mfpt2koff(mfpt_mle) 53 # samples /home/marscher/anaconda/envs/clean_ipy/lib/python2.7/site-packages/pyEMMA-2.1.1+89.ga553b9a-py2.7-linux-x86_64.egg/pyemma/msm/models/msm.pycin mfpt(self,A,B) 445 set of target states 446 """ --> 447 return self._mfpt(self.transition_matrix, A, B, mu=self.stationary_distribution) 448 449 def _committor_forward(self, P, A, B): /home/marscher/anaconda/envs/clean_ipy/lib/python2.7/site-packages/pyEMMA-2.1.1+89.ga553b9a-py2.7-linux-x86_64.egg/pyemma/msm/models/msm.pyc in _mfpt(self, P, A, B, mu) 433 from msmtools.analysis import mfpt as __mfpt 434 # scale mfpt by lag time --> 435 return self._timeunit_model.dt \* __mfpt(P, B, origin=A, mu=mu) 436 437 def mfpt(self, A, B): /home/marscher/workspace/msmtools/msmtools/analysis/api.pyc in mfpt(T, target, origin, tau, mu) 714 t_tau = dense.mean_first_passage_time.mfpt(T, target) 715 else: --> 716 t_tau = dense.mean_first_passage_time.mfpt_between_sets(T, target, origin, mu=mu) 717 718 # scale answer by lag time used. /home/marscher/workspace/msmtools/msmtools/analysis/dense/mean_first_passage_time.py in mfpt_between_sets(T, target, origin, mu) 128 129 """Mean first-passage time to Y (for all possible starting states)""" --> 130 tY = mfpt(T, target) 131 132 """Mean first-passage timefrom X toY""" /home/marscher/workspace/msmtools/msmtools/analysis/dense/mean_first_passage_time.py in mfpt(T, target) 87 b = np.ones(dim) 88 b[target] = 0.0 ---> 89 m_t = solve(A, b) 90 return m_t 91 /home/marscher/anaconda/envs/clean_ipy/lib/python2.7/site-packages/scipy/linalg/basic.pyc in solve(a, b, sym_pos, lower, overwrite_a, overwrite_b, debug, check_finite) 101 return x 102 if info > 0: --> 103 raise LinAlgError("singular matrix") 104 raise ValueError('illegal value in %d-th argument of internal gesv|posv' % 105 -info) LinAlgError: singular matrix — You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/markovmodel/PyEMMA/issues/786
Prof. Dr. Frank Noe Head of Computational Molecular Biology group Freie Universitaet Berlin
Phone: (+49) (0)30 838 75354 Web: research.franknoe.de
marscher
commented
8 years ago I've updated the notebooks, they are currently being executed in the jenkins job.
During testing the trypsin_benzamidine_hmm notebook I've encounter the following problem. For lag time=1, the evaluation of kinetic/thermodynamic quantities the mfpt method fails for matrix A being singular. However T is fully connected and reversible.
@nsplattner suggested that the new bhmm version gives a converged estimate, which is in this case not converged because we only feed in a very small data set.
We could handle the exception in the notebook, but I'd prefer a handler in the MSM.mfpt function (just to give an informative message, when we know how this is caused).
In the notebook cell 22: