TrajEnsemble#

class TrajEnsemble(trajs, tops=None, backend='no_load', common_str=None, basename_fn=None, traj_nums=None, custom_top=None)[source]#

A fancy list of single trajectories. Topologies can be different across trajs.

Check out http://statisticalbiophysicsblog.org/?p=92 for why trajectory ensembles are awesome.

This class is a fancy list of encodermap.trajinfo.info_single.SingleTraj`. Trajectories can have different topologies and will be grouped by the common_str argument. Each trajectory has its own unique traj_num, which identifies it in the ensemble - even when the ensemble is sliced or subsampled.

Examples

>>> import encodermap as em
>>> traj1 = em.SingleTraj.from_pdb_id("1YUG")
>>> traj2 = em.SingleTraj.from_pdb_id("1YUF")

Addition of two encodermap.trajinfo.info_single.SingleTraj also creates an ensemble.

>>> trajs = traj1 + traj2
>>> trajs  
<encodermap.TrajEnsemble object. Current backend is no_load. Containing 2 trajectories. Common str is ['1YUG', '1YUF']. Not containing any CVs...>

Indexing a TrajEnsemble returns a encodermap.trajinfo.info_single.SingleTraj based on its 0-based index. Think of the TrajEnsmeble as a list of encodermap.trajinfo.info_single.SingleTraj. But trajectories can also have traj_nums, which do not have to adhere to [0, 1, 2, ...]. This is similar to how a pandas.DataFrame offers indexing via .loc[] and .iloc[] (https://pandas.pydata.org/docs/user_guide/indexing.html#different-choices-for-indexing). For indexing trajs based on their traj_num, you can use the .tsel[] accessor of the TrajEnsmeble

Examples

>>> import encodermap as em
>>> traj1 = em.SingleTraj.from_pdb_id("1YUG")
>>> traj2 = em.SingleTraj.from_pdb_id("1YUF")

Addition of two SingleTraj also creates an ensemble.

>>> trajs = traj1 + traj2
>>> trajs.traj_nums
[0, 1]

Change the traj_num of traj2

>>> trajs[1].traj_num = 4
>>> trajs.traj_nums
[0, 4]
>>> trajs[1]  
<encodermap.SingleTraj object. Currently not in memory. Basename is '1YUF'. Not containing any CVs. Common string is '1YUF'. Object at ...>
>>> trajs.tsel[4]  
<encodermap.SingleTraj object. Currently not in memory. Basename is '1YUF'. Not containing any CVs. Common string is '1YUF'. Object at ...>

TrajEnsemble supports fancy indexing. You can slice to your liking (trajs[::5] returns a TrajEnsemble object that only consideres every fifth frame). Besides indexing by slices and integers, you can pass a 2-dimensional numpy.ndarray. np.array([[0, 5], [1, 10], [5, 20]]) will return a TrajEnsemble object with frame 5 of trajectory 0, frame 10 of trajectory 1 and frame 20 of trajectory 5.

Examples

>>> import encodermap as em
>>> traj1 = em.SingleTraj.from_pdb_id("1YUG")
>>> traj2 = em.SingleTraj.from_pdb_id("1YUF")
>>> trajs = traj1 + traj2
>>> sel = trajs[[[0, 0], [0, 1], [0, 2], [1, 10]]]
>>> sel  
<encodermap.TrajEnsemble object. Current backend is no_load. Containing 4 frames and 2 trajectories. Common str is...>

The TrajEnsemble class also is an iterator to iterate over trajectores. Besides plain iteration, the TrajEnsmeble also offers alternate iterators. The itertrajs() iterator returns a two-tuple of traj_num and traj. The iterframes() iterator returns a three-tuple of traj_num, frame_num, and traj.

Examples

>>> import encodermap as em
>>> traj1 = em.SingleTraj.from_pdb_id("1YUG")
>>> traj2 = em.SingleTraj.from_pdb_id("1YUF")
>>> trajs = traj1 + traj2
>>> trajs[1].traj_num = 4
>>> for traj_num, traj in trajs.itertrajs():
...     print(traj_num, traj.n_frames)
0 15
4 16
>>> for traj_num, frame_num ,traj in trajs.subsample(10).iterframes():
...     print(traj_num, frame_num, traj.n_frames)
0 0 1
0 10 1
4 0 1
4 10 1

The TrajEnsemble has multiple alternative constructors. The with_overwrite_trajnums constructor fixes inhomogeneous sequences of encodermap.trajinfo.info_single.SingleTraj and TrajEnsemble.

Examples

>>> import encodermap as em
>>> traj1 = em.SingleTraj.from_pdb_id("1YUG", traj_num=0)
>>> traj2 = em.SingleTraj.from_pdb_id("1YUF", traj_num=0)
>>> trajs = em.TrajEnsemble([traj1, traj2])  
Traceback (most recent call last):
    ...
Exception: The `traj_num` attributes of the provided 2 `SingleTraj`s is not unique, the `traj_num` 0 occurs 2 times. This can happen, if you use `SingleTraj`s, that are already part of a `TrajEnsemble`. To create copies of the `SingleTraj`s and overwrite their `traj_num`s, use the `with_overwrite_trajnums()` constructor.
>>> trajs = em.TrajEnsemble.with_overwrite_trajnums(traj1, traj2)
>>> trajs  
<encodermap.TrajEnsemble...>

The from_dataset constructor can be used to load an ensemble from an .h5 file

Examples

>>> import encodermap as em
>>> from tempfile import TemporaryDirectory
>>> traj1 = em.SingleTraj.from_pdb_id("1YUG")
>>> traj2 = em.SingleTraj.from_pdb_id("1YUF")
>>> trajs = em.TrajEnsemble([traj1, traj2])
>>> with TemporaryDirectory() as td:
...     trajs.save(td + "/trajs.h5")
...     new = em.TrajEnsemble.from_dataset(td + "/trajs.h5")
...     print(new)  
encodermap.TrajEnsemble object. Current backend is no_load. Containing 2 trajectories. Common str is...Not containing any CVs.

CVs#

The collective variables of the SingleTraj classes. Only CVs with matching names in all SingleTraj classes are returned. The data is stacked along a hypothetical time axis along the trajs.

Type:: dict[str, np.ndarray]

_CVs#

The same data as in CVs but with labels. Additionally, the xarray is not stacked along the time axis. It contains an extra dimension for trajectories.

Type:: xarray.Dataset

n_trajs#

Number of individual trajectories in this class.

Type:: int

n_frames#

Number of frames, sum over all trajectories.

Type:: int

locations#

A list with the locations of the trajectories.

Type:: list[str]

top#

A list with the reference pdb for each trajecotry.

Type:: list[mdtraj.Topology]

basenames#

A list with the names of the trajecotries. The leading path and the file extension is omitted.

Type:: list[str]

name_arr#

An array with len(name_arr) == n_frames. This array keeps track of each frame in this object by identifying each frame with a filename. This can be useful, when frames are mixed inside a TrajEnsemble class.

Type:: np.ndarray

Instantiate the TrajEnsmeble class with two lists of files.

Parameters:

(Union[Sequence[str] (trajs) – Sequence[SingleTraj], Sequence[Path]]): List of strings with paths to trajectories. Can also be a list of md.Trajectory or em.SingleTraj.
Sequence[md.Trajectory] – Sequence[SingleTraj], Sequence[Path]]): List of strings with paths to trajectories. Can also be a list of md.Trajectory or em.SingleTraj.
trajs (Union[Sequence[str], Sequence[Path], Sequence[md.Trajectory], Sequence[SingleTraj]])
tops (Union[None, Sequence[str], Sequence[Path]])
backend (Literal['mdtraj', 'no_load'])
common_str (Optional[Sequence[str]])
basename_fn (Optional[Callable[[str], str]])
traj_nums (Optional[Sequence[int]])
custom_top (Optional[CustomAAsDict])

:paramSequence[SingleTraj], Sequence[Path]]): List of strings with: paths to trajectories. Can also be a list of md.Trajectory or em.SingleTraj.

Parameters:

tops (Optional[list[str]]) – List of strings with paths to reference pdbs.
backend (str, optional) –
Choose the backend to load trajectories:
- ’mdtraj’ uses mdtraj, which loads all trajectories into RAM.
- ’no_load’ creates an empty trajectory object.
Defaults to ‘no_load’, which makes the instantiation of large ensembles fast and RAM efficient.
common_str (list[str], optional) – If you want to include trajectories with different topology. The common string is used to pair traj-files (.xtc, .dcd, .lammpstrj, ...) with their topology (.pdb, .gro, ...). The common-string should be a substring of matching traj and topology files.
basename_fn (Union[None, Callable[[str], str], optional) – A function to apply to the trajectory file path string to return the basename of the trajectory. If None is provided, the filename without extension will be used. When all files are named the same and the folder they’re in defines the name of the trajectory, you can supply lambda x: split('/')[-2] as this argument. Defaults to None.
custom_top (Optional[CustomAAsDict]) – Optional[CustomAAsDict]: An instance of the encodermap.trajinfo.trajinfo_utils.CustomTopology or a dictionary that can be made into such.
trajs (Union[Sequence[str], Sequence[Path], Sequence[md.Trajectory], Sequence[SingleTraj]])
traj_nums (Optional[Sequence[int]])

batch_iterator(batch_size: int, replace: bool = False, CV_names: tuple[str] = ('',), deterministic: bool = True, yield_index: bool = True, start: int = 1) → Iterator[tuple[ndarray, ndarray]][source]#

batch_iterator(batch_size: int, replace: bool = False, CV_names: tuple[str] = ('',), deterministic: bool = True, yield_index: bool = False, start: int = 1) → Iterator[ndarray]

batch_iterator(batch_size: int, replace: bool = False, CV_names: Sequence[str] | None = None, deterministic: bool = True, yield_index: bool = True, start: int = 1) → Iterator[tuple[ndarray, tuple[ndarray, ndarray, ndarray, ndarray, ndarray]]]

batch_iterator(batch_size: int, replace: bool = False, CV_names: Sequence[str] | None = None, deterministic: bool = True, yield_index: bool = False, start: int = 1) → Iterator[tuple[ndarray, ndarray, ndarray, ndarray, ndarray]]

Lazy batched iterator of CV data.

This iterator extracts batches of CV data from the ensemble. If the ensemble is a large HDF5 datset, this provides the ability to use all data without loading it all into memory.

Examples

Import EncoderMap and load some example trajectories.

>>> import encodermap as em
>>> trajs = em.TrajEnsemble(
...     [
...         'https://files.rcsb.org/view/1YUG.pdb',
...         'https://files.rcsb.org/view/1YUF.pdb'
...     ]
... )

This iterator will yield new samples forever. The batch is a tuple of numpy.ndarray.

>>> for batch in trajs.batch_iterator(batch_size=2):
...     print([b.shape for b in batch])
...     break
[(2, 148), (2, 147), (2, 150, 3), (2, 149), (2, 82)]

Use it with Python’s builtin next() function. The deterministic flag returns deterministic batches. The yield_index flag also provides the index of the extracted batch. In this example, both batches are extracted from the 1YUG trajectory (traj_num==0).

>>> iterator = trajs.batch_iterator(deterministic=True, batch_size=2, yield_index=True)
>>> index, batch = next(iterator)
>>> index
[[0 5]
 [0 8]]
>>> index, batch = next(iterator)
>>> index
[[ 0  3]
 [ 0 10]]

If a single string is requested for CV_names, the batch, will be a sinlge numpy.ndarray, rather than a tuple thereof.
>>> iterator = trajs.batch_iterator(batch_size=2, CV_names=["central_dihedrals"])

>>> batch = next(iterator)
>>> batch.shape
(2, 147)

Parameters:

batch_size (int) – The size of the batch.
replace (bool) – Whether inside a single batch a sample can occur more than once. Set to False (default) to only allow unique samples in a batch.
CV_names (Sequence[str]) – The names of the CVs to be used in the iterator. If a list/tuple with a single string is provided, the batch will be a numpy.ndarray, rather than a tuple thereof.
deterministic (bbol) – Whether the samples should be deterministic.
yield_index (bool) – Whether to also yield the index of the extracted samples.
start (int) – A start ineteger, which can be used together with deterministic=True to get different deterministic datasets.

Returns:

Different iterators based on chosen arguments.

Return type:

Iterator[Any]

cluster(cluster_id, col='cluster_membership', memberships=None, n_points=-1, overwrite=True)[source]#

Clusters this TrajEnsemble based on the provided cluster_id and col.

With ‘clustering’ we mean to extract a subset given a certain membership. Take two trajectories with 3 frames each as an ensemble. Let’s say we calculate the end-to-end distance of the trajectories and use it as a collective variable of the system. The values are [0.8, 1.3, 1.2, 1.9, 0.2, 1.3]. Based on these values, we define a boolean CV (using 0 as False and 1 as True) which says whether the end-to-end distance is smaller or grather than 1.0. We give this CV the name 'end_to_end_binary' and the values are [0, 1, 1, 1, 0, 1]. We can use this CV to ‘cluster’ the TrajEnsemble via:

cluster = trajs.cluster(cluster_id=0, col='end_to_end_binary'):
This gives a TrajEnsemble with 2 frames.

cluster = trajs.cluster(cluster_id=0, col='end_to_end_binary'):
This gives a TrajEnsemble with 4 frames.

Sometimes, you want to save this a cluster in a format that can be rendered by graphical programs (.xtc, .pdb), you can use either the join or stack method of the resulting :obj:``TrajEnsemble` to get a mdtraj.Trajectory, which is either stacked along the atom axis or joined along the time axis.

Note

If the resulting TrajEnsemble has inhomogeneous topologies, the join method will return a dict[md.Topology, md.Trajectory] instead. This dict can be used to save multiple (.xtc, .pdb) files and visualize your cluster in external programs.

The col parameter takes any CV name, that is per-frame and integer.

Parameters:

cluster_id (int) – The cluster id to use. Needs to be an integer, that is present in the col parameter.
col (str) – Which ‘column’ of the collective variables to use. Needs to be a key, that can be found in trajs.CVs.keys().
memberships (Optional[np.ndarray]) – If a numpy.ndarray is provided here, the memberships from this array will be used. In this case, the col argument will be unused.
n_points (int) – How many points the resulting cluster should contain. Subsamples the points in col == cluster_id evenly and without repeat. If set to -1, all points will be used.
overwrite (bool) – When the memberships argument is used, but the TrajEnsemble already has a CV under the name specified by col, you can set this to True to overwrite this column. Can be helpful, when you iteratively conduct multiple clusterings.

Return type:

TrajEnsemble

Examples

Import EncoderMap and NumPy.

>>> import encodermap as em
>>> import numpy as np

Load an example project.

>>> trajs = em.load_project("pASP_pGLU", load_autoencoder=False)

Create an array full of -1’s. These are the ‘outliers’.

>>> cluster_membership = np.ones(shape=(trajs.n_frames, )) * -1

Select the first 5 frames of every traj to be in cluster 0.

>>> cluster_membership[trajs.id[:, 1] < 5] = 0

Select all frames between 50 and 55 to be cluster 1.

>>> cluster_membership[(50 <= trajs.id[:, 1]) & (trajs.id[:, 1] <= 55)] = 1
>>> np.unique(cluster_membership)
array([-1.,  0.,  1.])

Load this array as a CV called 'clu_mem'.

>>> trajs.load_CVs(cluster_membership, attr_name='clu_mem')

Extract all of cluster 0 with n_points=-1.

>>> clu0 = trajs.cluster(0, "clu_mem")
>>> clu0.n_frames
35

Extract an evenly spaced subset of cluster 1 with 10 total points.

>>> clu1 = trajs.cluster(1, "clu_mem", n_points=10)
>>> clu1.n_frames
10

Cclusters with inhomogeneous topologies can be stacked along the atom axis.

>>> [t.n_atoms for t in trajs]
[69, 83, 103, 91, 80, 63, 73]
>>> stacked = clu1.stack()
>>> stacked.n_atoms
795

But joining the trajectories returns a dict[top, traj] if the topologies are inhomogeneous.

>>> joined = clu1.join()
>>> type(joined)
<class 'dict'>

dash_summary()[source]#

A pandas.DataFrame that summarizes this ensemble.

Returns:: The DataFrame.
Return type:: pd.DataFrame

del_CVs(CVs=None)[source]#

Deletes all CVs in all trajs. Does not affect the files.

Parameters:: CVs (Sequence[str] | None)
Return type:: None

del_featurizer()[source]#

Deletes the current instance of self.featurizer.

Return type:: None

classmethod from_textfile(fname, basename_fn=None)[source]#

Creates a TrajEnsemble object from a textfile.

The textfile needs to be space-separated with two or three columns:

Column 1:
The trajectory file.
Column 2:
The corresponding topology file (If you are using .h5 trajs, column 1 and 2 will be identical, but column 2 needs to be there nonetheless).
Column 3:
The common string of the trajectory. This column can be left out, which will result in an TrajEnsemble without common strings.

Parameters:

fname (Union[str, Path]) – File to be read.
basename_fn (Union[None, Callable[[str], str]], optional) – A function to apply to the traj_file string to return the basename of the trajectory. If None is provided, the filename without extension will be used. When all files are named the same and the folder they’re in defines the name of the trajectory, you can supply lambda x: split('/')[-2] as this argument. Defaults to None.

Returns:

A TrajEnsemble instance.

Return type:

TrajEnsemble

get_single_frame(key)[source]#

Returns a single frame from all loaded trajectories.

Consider a TrajEnsemble class with two trajectories. One has 10 frames, the other 5 (trajs.n_frames is 15). Calling trajs.get_single_frame(12) is equal to calling trajs[1][1]. Calling trajs.get_single_frame(16) will error, and trajs.get_single_frame(1) is the same as trajs[0][1].

Parameters:: key (int) – The frame to return.
Returns:: The frame.
Return type:: encodermap.trajinfo.info_single.SingleTraj

iterframes()[source]#

Generator over the frames in this instance.

Yields:

tuple –

A tuple containing the following:

int: The traj_num
int: The frame_num
encodermap.SingleTraj: An SingleTraj object.

Return type:

Iterator[tuple[int, int, SingleTraj]]

Examples

Import EncoderMap and load an example TrajEnsemble.

>>> import encodermap as em
>>> trajs = em.TrajEnsemble(
...     [
...         'https://files.rcsb.org/view/1YUG.pdb',
...         'https://files.rcsb.org/view/1YUF.pdb',
...     ],
... )
>>> print(trajs.n_frames)
31

Subsample every tenth frame.

>>> trajs = trajs.subsample(10)
>>> trajs.n_frames
4

Call the iterframes method.

>>> for traj_num, frame_num, frame in trajs.iterframes():
...     print(traj_num, frame_num, frame.n_frames)
0 0 1
0 10 1
1 0 1
1 10 1

itertrajs()[source]#

Generator over the SingleTraj classes.

Yields:

tuple –

A tuple containing the following:

int: A loop-counter integer. Is identical with traj.traj_num.
encodermap.SingleTraj: An SingleTraj object.

Return type:

Iterator[tuple[int, SingleTraj]]

Examples

>>> import encodermap as em
>>> trajs = em.TrajEnsemble(
...     [
...         'https://files.rcsb.org/view/1YUG.pdb',
...         'https://files.rcsb.org/view/1YUF.pdb'
...     ]
... )
>>> for i, traj in trajs.itertrajs():
...     print(traj.basename)
1YUG
1YUF

load_CVs(data=None, attr_name=None, cols=None, deg=None, periodic=True, labels=None, directory=None, ensemble=False, override=False, custom_aas=None, alignment=None)[source]#

Loads CVs in various ways. The easiest way is to provide a single numpy.ndarray and a name for that array.

Besides np.ndarray, files (.txt and .npy) can be loaded. Features or Featurizers can be provided. A xarray.Dataset can be provided. A str can be provided which either is the name of one of EncoderMap’s features (encodermap.features) or the string can be ‘all’, which loads all features required for EncoderMap’s encodermap.autoencoder.autoencoder`AngleDihedralCartesianEncoderMap.

Parameters:

data (Optional[TrajEnsembleFeatureType]) – The CV to load. When a numpy.ndarray is provided, it needs to have a shape matching n_frames and the data will be distributed to the trajs, When a list of files is provided, len(data) (the files) needs to match n_trajs. The first file will be loaded by the first traj (based on the traj’s traj_num) and so on. If a list of numpy.ndarray is provided, the first array will be assigned to the first traj (based on the traj’s traj_num). If None is provided, the argument directory will be used to construct a str using this expression fname = directory + traj.basename + '_' + attr_name. If there are .txt or .npy files matching that string in the directory, the CVs will be loaded from these files to the corresponding trajs. Defaults to None.
attr_name (Optional[str]) – The name under which the CV should be found in the class. Choose whatever you like. 'highd', 'lowd', 'dists', etc. The CV can then be accessed via dot-notation: trajs.attr_name. Defaults to None, in which case, the argument data should point to existing files. The attr_name will be extracted from these files.
cols (Optional[list[int]]) –
A list of integers indexing the columns of the data to be loaded. This is useful if a file contains columns which are not features (i.e. an indexer or the error of the features. eg:
```
id   f1    f2    f1_err    f2_err
0    1.0   2.0   0.1       0.1
1    2.5   1.2   0.11      0.52
```
In that case, you would want to supply cols=[1, 2] to the cols argument. If None is provided all columns are loaded. Defaults to None.
deg (Optional[bool]) – Whether to return angular CVs using degrees. If None or False, CVs will be in radian. Defaults to None.
periodic (bool) – Whether to use the minimum image convention to calculate distances/angles/dihedrals. This is generally recommended, when you don’t clean up your trajectories and the proteins break over the periodic boundary conditions. However, when the protein is large, the distance between one site and another might be shorter through the periodic boundary. This can lead to wrong results in your distance calculations.
labels (list[str]) – A list containing the labels for the dimensions of the data. If you provide a numpy.ndarray with shape (n_trajs, n_frames, n_feat), this list needs to be of len(n_feat). An exception will be raised otherwise. If None is privided, the labels will be automatically generated. Defaults to None.
directory (Optional[str]) – If this argument is provided, the directory will be searched for .txt or .npy files which have the same names as the trajectories have basenames. The CVs will then be loaded from these files.
ensemble (bool) – Whether the trajs in this class belong to an ensemble. This implies that they contain either the same topology or are very similar (think wt, and mutant). Setting this option True will try to match the CVs of the trajs onto the same dataset. If a VAL residue has been replaced by LYS in the mutant, the number of sidechain dihedrals will increase. The CVs of the trajs with VAL will thus contain some NaN values. Defaults to False.
override (bool) – Whether to override CVs with the same name as attr_name.
custom_aas (Optional[CustomAAsDict]) – You can provide non-standard residue definitions in this argument. See encodermap.trajinfo.trajinfo_utils.CustomTopology for information how to use the custom_aas argument. If set to None (default), only standard residue names are assumed.
alignment (Optional[str]) – If your proteins have similar but different sequences, you can provide a CLUSTAL W alignment as this argument and the featurization will align the features accordingly.

Raises:

TypeError – When wrong Type has been provided for data.

Return type:

None

load_custom_topology(custom_top=None)[source]#

Loads a custom_topology from a CustomTopology class or a dict.

See also

CustomTopology

Parameters:: custom_top (CustomTopology | dict[str | tuple[str, str], None | tuple[str, None] | tuple[str, dict[Literal['bonds', 'optional_bonds', 'delete_bonds', 'optional_delete_bonds', 'PHI', 'PSI', 'OMEGA', 'not_PHI', 'not_PSI', 'not_OMEGA', 'CHI1', 'CHI2', 'CHI3', 'CHI4', 'CHI5'], list[str] | list[tuple[str | int, str | int]]]]] | None) – Optional[Union[CustomTopology, CustomAAsDict]]: An instance of the CustomTopology class or a dictionary that can be made into such.
Return type:: None

load_trajs()[source]#

Loads all trajs in self.

Return type:: None

parse_clustal_w_alignment(aln)[source]#

Parse an alignment in ClustalW format and add the info to the trajectories.

Parameters:: aln (str) – The alignment in ClustalW format.
Return type:: None

save(fname, CVs='all', overwrite=False, only_top=False)[source]#

Saves this TrajEnsemble into a single .h5 file.

Parameters:

fname (Union[str, Path]) – Where to save the file.
CVs (Union[Literal["all"], list[str], Literal[False]]) – Which CVs to alos store in the file. If set to 'all', all CVs will be saved. Otherwise, a list[str] can be provided to only save specific CVs. Can also be set to False, no CVs are stored in the file.
overwrite (bool) – If the file exists, it is overwritten.
only_top (bool) – Only writes the trajectorie’s topologies into the file.

Raises:

IOError – If file already exists and overwrite is not True.

Return type:

None

save_CVs(path)[source]#

Saves the CVs to a NETCDF file using xarray.

Parameters:: path (str | Path)
Return type:: None

sidechain_info()[source]#

Indices used for the AngleDihedralCartesianEncoderMap class to allow training with multiple different sidechains.

Returns:: The indices. The key ‘-1’ is used for the hypothetical convex hull of all feature spaces (the output of the tensorflow model). The other keys match the common_str of the trajs.
Return type:: dict[str, Sequence[int]]
Raises:: Exception – When the common_strings and topologies are not aligned. An exception is raised. Aligned means that all trajs with the same common_str should possess the same topology.

split_into_frames(inplace=False)[source]#

Splits self into separate frames.

Parameters:: inplace (bool) – Whether to do the split inplace or not. Defaults to False and thus, returns a new TrajEnsemble class.
Return type:: None

subsample(stride=None, total=None)[source]#

Returns a subset of this TrajEnsemble given the provided stride or total.

This is a faster alternative than using the trajs[trajs.index_arr[::1000]] when HDF5 trajs are used, because the slicing information is saved in the respective encodermap.trajinfo.info_single.SingleTraj

and loading of single frames is faster in HDF5 formatted trajs.

Parameters:

stride (Optional[int]) – Return a frame ever stride frames.
total (Optional[int]) – Return a total of evenly sampled frames.

Returns:

A trajectory ensemble.

Return type:

TrajEnsemble

Note

The result from subsample(1000)` `is different from ``trajs[trajs.index_arr[::1000]]. With subsample every trajectory is sub-sampled independently. Consider a TrajEnsemble with two encodermap.trajinfo.info_single.SingleTraj trajectories with 18 frames each. subsampled = trajs.subsample(5) would return a TrajEnsemble with two trajs with 3 frames each (subsampled.n_frames == 6). Whereas, subsampled = trajs[trajs.index_arr[::5]] would return a TrajEnsemble with 7 SingleTrajs with 1 frame each (subsampled.n_frames == 7). Because the time and frame numbers are saved all the time, this should not be too much of a problem.

to_alignment_query()[source]#

A string, that cen be put into sequence alignment software.

Return type:: str

unload()[source]#

Unloads all trajs in self.

Return type:: None

classmethod with_overwrite_trajnums(*trajs)[source]#

Creates a TrajEnsemble by copying the provided encodermap.trajinfo.info_single.SingleTraj instances and changing their traj_num attribute to adhere to [0, 1, 2, ...].

Parameters:: trajs (Sequence[SingleTraj]) – The sequence of trajs.
Returns:: A TrajEnsemble instance.
Return type:: TrajEnsemble

property CVs: dict[str, ndarray]#

Returns dict of CVs in SingleTraj classes. Only CVs with the same names in all SingleTraj classes are loaded.

Type:: dict

property CVs_in_file: bool#

Is true, if CVs can be loaded from file. Can be used to build a data generator from.

Type:: bool

property _CVs: Dataset#

Returns x-array Dataset of matching CVs. stacked along the trajectory-axis.

Type:: xarray.Dataset

property basenames: list[str]#

List of the basenames in the Info single classes.

Type:: list

property frames: list[int]#

Frames of individual trajectories.

Type:: list

property id: ndarray#

Duplication of self.index_arr

Type:: np.ndarray

property index_arr: ndarray#

Returns np.ndarray with ndim = 2. Clearly assigning every loaded frame an identifier of traj_num (self.index_arr[:,0]) and frame_num (self.index_arr[:,1]). Can be used to create an unspecified subset of frames and can be useful when used with clustering.

Type:: np.ndarray

property locations: list[str]#

Duplication of self.traj_files but using the trajs own traj_file attribute. Ensures that traj files are always returned independent of the current load state.

Type:: list

property n_frames: int#

Sum of the loaded frames.

Type:: int

property n_residues: int#

List of number of residues of the SingleTraj classes

Type:: list

property n_trajs: int#

Number of trajectories in this ensemble.

Type:: int

property name_arr: ndarray#

Trajectory names with the same length as self.n_frames.

Type:: np.ndarray

property top: list[Topology]#

Returns a minimal set of mdtraj.Topologies.

If all trajectories share the same topology a list with len 1 will be returned.

Type:: list

property top_files: list[str]#

Returns minimal set of topology files.

If yoy want a list of top files with the same length as self.trajs use self._top_files and self._traj_files.

Type:: list

property traj_files: list[str]#

A list of the traj_files of the individual SingleTraj classes.

Type:: list

property traj_joined: Trajectory#

Returns a mdtraj Trajectory with every frame of this class appended along the time axis.

Can also work if different topologies (with the same number of atoms) are loaded. In that case, the first frame in self will be used as topology parent and the remaining frames’ xyz coordinates are used to position the parents’ atoms accordingly.

Examples

>>> import encodermap as em
>>> trajs = em.load_project("pASP_pGLU")
>>> subsample = trajs[0][:20] + trajs[1][:20]
>>> subsample.split_into_frames().traj_joined  
<mdtraj.Trajectory with 40 frames, 69 atoms, 6 residues, and unitcells at ...>

Type:: mdtraj.Trajectory

property traj_nums: list[int]#

Number of info single classes in self.

Type:: list

property trajs_by_common_str: dict[None | str, TrajEnsemble]#

Returns the trajs in self ordered by top.

If all trajectories share the same common_str, a dict with one key will be returned. As the common_str can be None, None can also occur as a key in this dict.

Type:: dict[str, TrajEnsemble]

property trajs_by_top: dict[Topology, TrajEnsemble]#

Returns the trajs in self ordered by top.

If all trajectories share the same topology, a dict with one key will be returned.

Type:: dict[md.Topology, TrajEnsemble]

property xyz: ndarray#

xyz coordinates of all atoms stacked along the traj-time axis.

Only works if all trajs share the same topology.

Type:: np.ndarray

TrajEnsemble#

This Page