/deepchem

Democratizing Deep-Learning for Drug Discovery, Quantum Chemistry, Materials Science and Biology

Primary LanguagePythonMIT LicenseMIT

DeepChem

Build Status Coverage Status Anaconda-Server Badge PyPI version

DeepChem aims to provide a high quality open-source toolchain that democratizes the use of deep-learning in drug discovery, materials science, quantum chemistry, and biology.

Table of contents:

Requirements

Installation

deepchem currently supports both Python 2.7 and Python 3.5, and is supported on 64 bit Linux and Mac OSX. Please make sure you follow the directions below precisely. While you may already have system versions of some of our dependencies, there is no guarantee that deepchem will work with alternate versions than those specified below.

Note that when using Ubuntu 16.04 server or similar environments, you may need to ensure libxrender is provided via e.g.:

sudo apt-get install -y libxrender-dev

Using a conda environment

You can install deepchem in a new conda environment using the conda commands in scripts/install_deepchem_conda.sh Installing via this script will ensure that you are installing from the source.

git clone https://github.com/deepchem/deepchem.git      # Clone deepchem source code from GitHub
cd deepchem
bash scripts/install_deepchem_conda.sh deepchem
source activate deepchem
yes | pip install tensorflow-gpu==1.6.0      # If you want GPU support
python setup.py install                                # Manual install
nosetests -a '!slow' -v deepchem --nologcapture        # Run tests

This creates a new conda environment deepchem and installs in it the dependencies that are needed. To access it, use the conda activate deepchem command (if your conda version >= 4.4) and use source activate deepchem command (if your conda version < 4.4).

Check this link for more information about the benefits and usage of conda environments. Warning: Segmentation faults can still happen via this installation procedure.

Easy Install via Conda

conda install -c deepchem -c rdkit -c conda-forge -c omnia deepchem=2.0.0

Note: Easy Install installs the latest stable version of deepchem and does not install from source. If you need to install from source make sure you follow the steps here.

Using a Docker Image

Using a docker image requires an NVIDIA GPU. If you do not have a GPU please follow the directions for using a conda environment In order to get GPU support you will have to use the nvidia-docker plugin.

# This will the download the latest stable deepchem docker image into your images
docker pull deepchemio/deepchem

# This will create a container out of our latest image with GPU support
nvidia-docker run -i -t deepchemio/deepchem

# You are now in a docker container whose python has deepchem installed
# For example you can run our tox21 benchmark
cd deepchem/examples
python benchmark.py -d tox21

# Or you can start playing with it in the command line
pip install jupyter
ipython
import deepchem as dc

FAQ

  1. Question: I'm seeing some failures in my test suite having to do with MKL Intel MKL FATAL ERROR: Cannot load libmkl_avx.so or libmkl_def.so.

    Answer: This is a general issue with the newest version of scikit-learn enabling MKL by default. This doesn't play well with many linux systems. See BVLC/caffe#3884 for discussions. The following seems to fix the issue

    conda install nomkl numpy scipy scikit-learn numexpr
    conda remove mkl mkl-service

Getting Started

Two good tutorials to get started are Graph Convolutional Networks and Multitask_Networks_on_MUV. Follow along with the tutorials to see how to predict properties on molecules using neural networks.

Afterwards you can go through other tutorials, and look through our examples in the examples directory. To apply deepchem to a new problem, try starting from one of the existing examples or tutorials and modifying it step by step to work with your new use-case. If you have questions or comments you can raise them on our gitter.

Input Formats

Accepted input formats for deepchem include csv, pkl.gz, and sdf files. For example, with a csv input, in order to build models, we expect the following columns to have entries for each row in the csv file.

  1. A column containing SMILES strings [1].
  2. A column containing an experimental measurement.
  3. (Optional) A column containing a unique compound identifier.

Here's an example of a potential input file.

Compound ID measured log solubility in mols per litre smiles
benzothiazole -1.5 c2ccc1scnc1c2

Here the "smiles" column contains the SMILES string, the "measured log solubility in mols per litre" contains the experimental measurement and "Compound ID" contains the unique compound identifier.

[2] Anderson, Eric, Gilman D. Veith, and David Weininger. "SMILES, a line notation and computerized interpreter for chemical structures." US Environmental Protection Agency, Environmental Research Laboratory, 1987.

Data Featurization

Most machine learning algorithms require that input data form vectors. However, input data for drug-discovery datasets routinely come in the format of lists of molecules and associated experimental readouts. To transform lists of molecules into vectors, we need to subclasses of DeepChem loader class dc.data.DataLoader such as dc.data.CSVLoader or dc.data.SDFLoader. Users can subclass dc.data.DataLoader to load arbitrary file formats. All loaders must be passed a dc.feat.Featurizer object. DeepChem provides a number of different subclasses of dc.feat.Featurizer for convenience.

Performances

In depth performance tables for DeepChem models are available on MoleculeNet.ai

Gitter

Join us on gitter at https://gitter.im/deepchem/Lobby. Probably the easiest place to ask simple questions or float requests for new features.

DeepChem Publications

  1. Computational Modeling of β-secretase 1 (BACE-1) Inhibitors using Ligand Based Approaches
  2. Low Data Drug Discovery with One-Shot Learning
  3. MoleculeNet: A Benchmark for Molecular Machine Learning
  4. Atomic Convolutional Networks for Predicting Protein-Ligand Binding Affinity

About Us

DeepChem is possible due to notable contributions from many people including Peter Eastman, Evan Feinberg, Joe Gomes, Karl Leswing, Vijay Pande, Aneesh Pappu, Bharath Ramsundar and Michael Wu (alphabetical ordering). DeepChem was originally created by Bharath Ramsundar with encouragement and guidance from Vijay Pande.

DeepChem started as a Pande group project at Stanford, and is now developed by many academic and industrial collaborators. DeepChem actively encourages new academic and industrial groups to contribute!

Corporate Supporters

DeepChem is supported by a number of corporate partners who use DeepChem to solve interesting problems.

Schrödinger

Schödinger

DeepChem has transformed how we think about building QSAR and QSPR models when very large data sets are available; and we are actively using DeepChem to investigate how to best combine the power of deep learning with next generation physics-based scoring methods.

DeepCrystal

DeepCrystal Logo

DeepCrystal was an early adopter of DeepChem, which we now rely on to abstract away some of the hardest pieces of deep learning in drug discovery. By open sourcing these efficient implementations of chemically / biologically aware deep-learning systems, DeepChem puts the latest research into the hands of the scientists that need it, materially pushing forward the field of in-silico drug discovery in the process.

Version

2.0.0