MolML

Build Status Coverage Status Documentation Status PyPI version License

A library to interface molecules and machine learning. The goal of this library is to be a simple way to convert molecules into a vector representation for later use with libraries such as scikit-learn. This is done using a similar API scheme.

All of the coordinates are assumed to be in angstroms.

Features

- Simple interface to many common molecular descriptors and their variants
    - Molecule
        - Coulomb Matrix
        - Bag of Bonds
        - Encoded Bonds
        - Encoded Angles
        - Connectivity
        - Connectivity Tree
        - Autocorrelation
    - Atom
        - Shell
        - Local Encoded Bonds
        - Local Encoded Angles
        - Local Coulomb Matrix
        - Behler-Parrinello
    - Kernel
        - Atom/Summation Kernel
    - Fragment
        - FragmentMap
    - Crystal
        - Generallized Crystal
        - Ewald Sum Matrix
        - Sine Matrix
- Parallel feature generation
- Ability to save/load fit models
- Multiple input formats supported (and ability to define your own)
- Supports both Python 2 and Python 3

Example Usage

    >>> from molml.features import CoulombMatrix
    >>> feat = CoulombMatrix()
    >>> H2 = (
    ...         ['H', 'H'],
    ...         [
    ...             [0.0, 0.0, 0.0],
    ...             [1.0, 0.0, 0.0],
    ...         ]
    ... )
    >>> HCN = (
    ...         ['H', 'C', 'N'],
    ...         [
    ...             [-1.0, 0.0, 0.0],
    ...             [ 0.0, 0.0, 0.0],
    ...             [ 1.0, 0.0, 0.0],
    ...         ]
    ... )
    >>> feat.fit([H2, HCN])
    CoulombMatrix(input_type='list', n_jobs=1, sort=False, eigen=False, drop_values=False, only_lower_triangle=False)
    >>> feat.transform([H2])
    array([[ 0.5,  1. ,  0. ,  1. ,  0.5,  0. ,  0. ,  0. ,  0. ]])
    >>> feat.transform([H2, HCN])
    array([[  0.5      ,   1.       ,   0.       ,   1.       ,   0.5      ,
            0.       ,   0.       ,   0.       ,   0.       ],
            [  0.5      ,   6.       ,   3.5      ,   6.       ,  36.8581052,
            42.       ,   3.5      ,  42.       ,  53.3587074]])
    >>>
    >>> # Example loading from files directly
    >>> feat2 = CoulombMatrix(input_type='filename')
    CoulombMatrix(input_type='filename', n_jobs=1, sort=False, eigen=False, drop_values=False, only_lower_triangle=False)
    >>> paths = ['data/qm7/qm-%04d.out' % i for i in xrange(2)]
    >>> feat2.fit_transform(paths)
    array([[ 36.8581052 ,   5.49459021,   5.49462885,   5.4945    ,
              5.49031286,   0.        ,   0.        ,   0.        ,
              5.49459021,   0.5       ,   0.56071947,   0.56071656,
              0.56064037,   0.        ,   0.        ,   0.        ,
              5.49462885,   0.56071947,   0.5       ,   0.56071752,
              0.56064089,   0.        ,   0.        ,   0.        ,
              5.4945    ,   0.56071656,   0.56071752,   0.5       ,
              0.56063783,   0.        ,   0.        ,   0.        ,
              5.49031286,   0.56064037,   0.56064089,   0.56063783,
              0.5       ,   0.        ,   0.        ,   0.        ,
              0.        ,   0.        ,   0.        ,   0.        ,
              0.        ,   0.        ,   0.        ,   0.        ,
              0.        ,   0.        ,   0.        ,   0.        ,
              0.        ,   0.        ,   0.        ,   0.        ,
              0.        ,   0.        ,   0.        ,   0.        ,
              0.        ,   0.        ,   0.        ,   0.        ],
           [ 36.8581052 ,  23.81043959,   5.48396427,   5.48394941,
              5.4837656 ,   2.78378686,   2.78375582,   2.78376439,
              23.8104396,  36.8581052 ,   2.78378953,   2.78375777,
              2.78375823,   5.4839846 ,   5.48393324,   5.48376877,
              5.48396427,   2.78378953,   0.5       ,   0.56363019,
              0.56362464,   0.40019757,   0.39971446,   0.3261774 ,
              5.48394941,   2.78375777,   0.56363019,   0.5       ,
              0.56362305,   0.39971429,   0.32617621,   0.40019524,
              5.4837656 ,   2.78375823,   0.56362464,   0.56362305,
              0.5       ,   0.32617702,   0.40019469,   0.3997145 ,
              2.78378686,   5.4839846 ,   0.40019757,   0.39971429,
              0.32617702,   0.5       ,   0.56362996,   0.56362587,
              2.78375582,   5.48393324,   0.39971446,   0.32617621,
              0.40019469,   0.56362996,   0.5       ,   0.56362278,
              2.78376439,   5.48376877,   0.3261774 ,   0.40019524,
              0.3997145 ,   0.56362587,   0.56362278,   0.5       ]])

For more examples, look in the examples. Note: To run some of the examples scikit-learn>=0.16.0 is required.

For the full documentation, refer to the docs or the docstrings in the code.

Dependencies

MolML works with both Python 2 and Python 3. It has been tested with the versions listed below, but newer versions should work.

python>=2.7/3.5/3.6
numpy>=1.9.1
scipy>=0.15.1
pathos>=0.2.0
bidict>=0.17.5
future  # For python 2

NOTE: Due to an issue with multiprocess (a pathos dependency), the minimum version of Python that will work is 2.7.4. For full details see this link. Without this, the parallel computation of features will fail.

Install

Once numpy and scipy are installed, the package can be installed with pip.

$ pip install molml

Or for the bleeding edge version, you can use

$ pip install git+git://github.com/crcollins/molml

Development

To install a development version, just clone the git repo.

$ git clone https://github.com/crcollins/molml
$ # cd to molml and setup some virtualenv
$ pip install -r requirements-dev.txt

Pull requests and bug reports are welcomed!

To build the documentation, you just need to install the documentation dependencies. These are already included in the dev install.

$ cd docs/
$ pip install -r requirements-docs.txt
$ make html

Testing

To run the tests, make sure that nose is installed and then run:

$ nosetests

To include coverage information, make sure that coverage is installed and then run:

$ nosetests --with-coverage --cover-package=molml --cover-erase

Citation

Currently, there is not a dedicated publication for MolML. Instead, feel free to cite the work that spawned this library.

@article{collins2018constant,
    title={Constant size descriptors for accurate machine learning models of molecular properties},
    author={Collins, Christopher R and Gordon, Geoffrey J and von Lilienfeld, O Anatole and Yaron, David J},
    journal={The Journal of Chemical Physics},
    volume={148},
    number={24},
    pages={241718},
    year={2018},
    publisher={AIP Publishing}
}

In addition, each feature extraction method has its own main reference listed in the docstring. These can also be accessed as follows:

    >>> from molml.features import CoulombMatrix
    >>> print(CoulombMatrix().get_citation())
    Rupp, M.; Tkatchenko, A.; Muller, K.-R.; von Lilienfeld, O. A. Fast and Accurate Modeling of Molecular Atomization Energies with Machine Learning. Phys. Rev. Lett. 2012, 108, 058301.
    Hansen, K.; Montavon, G.; Biegler, F.; Fazli, S.; Rupp, M.; Scheffler, M.; von Lilienfeld, O. A.; Tkatchenko, A.; Muller, K.-R. Assessment and Validation of Machine Learning Methods for Predicting Molecular Atomization Energies. J. Chem. Theory Comput. 2013, 9, 3404-3419.