tltsne

Time-lagged t-SNE of molecular trajectories.

Authors

Vojtech Spiwok and Pavel Kriz

Usage

Trajectory of molecular simulation is dimensionally reduced by t-distributed stochastic embedding (t-SNE) [1] and by a version of t-SNE that focuses on slow motions via analysis inspired by time-lagged independent component analysis (TICA) [2,3].

The input is a trajectory in pdb, xtc, trr, dcd, netcdf or mdcrd (only atoms intended for analysis). The second input file is a topology (pdb file with same atoms as in trajectory). Output contains frame ID, PCA, TICA, t-SNE and time-lagged t-SNE coordinates.

usage: tltsne [-h] [-i INFILE] [-p INTOP] [-o OUTFILE] [-nofit NOFIT]
              [-lagtime LAGTIME] [-pcadim PCADIM] [-ticadim TICADIM]
              [-maxpcs MAXPCS] [-ncomp NCOMP] [-perplex1 PERPLEX1]
              [-perplex2 PERPLEX2] [-rate RATE] [-niter NITER] [-exag EXAG]

Time-lagged t-SNE of simulation trajectories, requires scimpy, pyemma, sklearn
and mdtraj

optional arguments:
  -h, --help          show this help message and exit
  -i INFILE           Input trajectory in pdb, xtc, trr, dcd, netcdf or mdcrd
                      of atoms to be analyzed. No jumps in PBC allowed.
  -p INTOP            Input topology in pdb with atoms to be analyzed.
  -o OUTFILE          Output file.
  -nofit NOFIT        Structure is NOT fit to reference topology if nofit is
                      set to 1 (default 0).
  -lagtime LAGTIME    Lag time in number of frames (default 1).
  -pcadim PCADIM      Number o PCA coordinates to be printed (defaut 2).
  -ticadim TICADIM    Number o TICA coordinates to be printed (defaut 2).
  -maxpcs MAXPCS      Number of TICA coordinates to be passed to t-SNE
                      (default 50).
  -ncomp NCOMP        Number of t-SNE and time-lagged t-SNE coordinates to be
                      printed (defaut 2).
  -perplex1 PERPLEX1  Perplexity of t-SNE (default 10.0).
  -perplex2 PERPLEX2  Perplexity of time-lagged t-SNE (default 10.0).
  -rate RATE          Learnning rate of t-SNE and time-lagged t-SNE (default
                      200.0).
  -niter NITER        Number of iterations of t-SNE and time-lagged t-SNE
                      (default 1000).
  -exag EXAG          Early exaggeration of t-SNE and time-lagged t-SNE.

Install

Install via PIP:

pip3 install tltsne

(or with sudo).

Install from GitHub:

git clone https://github.com/spiwokv/tltsne.git
cd tltsne
pip3 install .

(or with sudo).

Thanks

• pyemma [4]
• mdtraj [5]
• scipy [6]
• sklearn [7]

References

1. L.J.P. van der Maaten, G.E. Hinton. Visualizing High-Dimensional Data Using t-SNE. J. Mach. Learn. Res. 2008, 9, 2579-2605.

2. G. Perez-Hernandez, F. Paul, T. Giorgino, G. de Fabritiis, F. Noé: Identification of slow molecular order parameters for Markov model construction. J. Chem. Phys. 2013, 139, 015102.

3. C. R. Schwantes and V. S. Pande: Improvements in Markov state model construction reveal many non-native interactions in the folding of NTL9. J. Chem. Theory Comput. 2013, 9, 2000-2009.

4. http://emma-project.org/

5. http://mdtraj.org/1.9.3/

6. https://www.scipy.org/

7. https://scikit-learn.org/