This repository is no longer being maintained and has been archived for historical purposes.
We have now developed DeepRank2, an improved and unified version of DeepRank-Mut, DeepRank, and DeepRank-GNN.
:sparkles: DeepRank2 allows for transformation and storage of 3D representations of both protein-protein interfaces (PPIs) and protein single-residue variants (SRVs) into either graphs or volumetric grids containing structural and physico-chemical information. These can be used for training neural networks for a variety of patterns of interest, using either our pre-implemented training pipeline for graph neural networks (GNNs) or convolutional neural networks (CNNs) or external pipelines.
We look forward to seeing you in our new space - DeepRank2!
DeepRank-mut is a general, configurable deep learning framework for predicting pathogenicity of missense variants using 3D convolutional neural networks (CNNs).
DeepRank-mut contains useful APIs for pre-processing protein structural data, computing features for atoms/residues surrounding the missense variant, as well as training and testing CNN models.
DeepRank requires a Python version 3.7 or 3.8 on Linux and MacOS.
You can also install the under development source code from the branch development
git clone --branch development https://github.com/DeepRank/DeepRank-mut.gitcd deeprank-mutpip install -e ./To check if installation is successful, you can run a test
pytestWe give here the tutorial like introduction to the DeepRank machinery. We quickly illsutrate here the three main steps of Deeprank-mut:
For data generation, PDB files must be stored locally. The user can optionally provide PSSM files, mapped to those PDB files. Together with the PDB files, variant data must be available. Here's an example of a table, containing variant data:
| PDB ID | CHAIN | RESIDUE | WILDTYPE | VARIANT | CLASS |
|---|---|---|---|---|---|
| 101m | A | 25 | GLY | ALA | BENIGN |
| 101m | A | 21 | VAL | SER | BENIGN |
The CLASS column would typically be omitted from unclassified variants.
In this example, we store the table in CSV format.
All the features/targets and mapped features onto grid points will be auomatically calculated and store in a HDF5 file.
import pandas
from deeprank.models.variant import *
from deeprank.generate import *
from mpi4py import MPI
from deeprank.domain.amino_acid import amino_acids_by_code
from deeprank.models.environment import Environment
comm = MPI.COMM_WORLD
# let's put this sample script in the test folder, so the working path will be ./test/
# name of the hdf5 to generate
hdf5_path = 'train_data.hdf5'
environment = Environment(pdb_root="path/to/pdb/files/",
pssm_root="path/to/pssm/files/")
table = pandas.read_csv("table.csv")
variants = []
for _, row in table.iterrows():
wildtype_amino_acid = amino_acids_by_code[row["WILDTYPE"]]
variant_amino_acid = amino_acids_by_code[row["VARIANT"]]
variant = PdbVariantSelection(
row["PDB ID"], row["CHAIN"], row["RESIDUE"],
wildtype_amino_acid=wildtype_amino_acid,
variant_amino_acid=variant_amino_acid,
variant_class=VariantClass.parse(row["CLASS"]), # variant class is optional
)
variants.append(variant)
# initialize the database
database = DataGenerator(environment, variants=variants,
data_augmentation=10,
compute_targets=['deeprank.targets.variant_class'], # omit this, if the variant class is not provided
compute_features=[
'deeprank.features.atomic_contacts',
'deeprank.features.neighbour_profile',
'deeprank.features.accessibility'],
hdf5=hdf5_path,
mpi_comm=comm)
# create the database
# compute features/targets for all complexes
database.create_database(prog_bar=True)
# define the 3D grid
grid_info = {
'number_of_points': [30,30,30],
'resolution': [1.,1.,1.],
'atomic_densities': {'C': 1.7, 'N': 1.55, 'O': 1.52, 'S': 1.8},
}
# Map the features
database.map_features(grid_info,try_sparse=True, time=False, prog_bar=True)
This script can be exectuted using for example 4 MPI processes with the command:
NP=4
mpiexec -n $NP python generate.pyIn the first part of the script we define the path where to find the PDBs of the native structure of interest. We then specify position of variant in the structure and the variant amino acid, to enable the code to compute features of the amino acids surrounding the variant.
We then initialize the DataGenerator object. This object (defined in deeprank/generate/DataGenerator.py) needs a few input parameters:
We then create the data base with the command database.create_database(). This function automatically creates an HDF5 files where the pdb has its own group. In each group we can find the pdb, its calculated features and the target. We can now mapped the features to a grid. This is done via the command database.map_features(). As you can see this method requires a dictionary as input. The dictionary contains the instruction to map the data.
The atomic densities are mapped following the protein-ligand paper. The other features are mapped to the grid points using a Gaussian function (other modes are possible but somehow hard coded)
To explore the HDf5 file and vizualize the features you can use the dedicated browser https://github.com/DeepRank/DeepXplorer. This tool allows to dig through the hdf5 file and to directly generate the files required to vizualize the features in VMD or PyMol. An iPython comsole is also embedded to analyze the feature values, plot them etc ....
The HDF5 files generated above can be used as input for deep learning experiments. You can take a look at the file test/test_learn.py for some examples. We give here a quick overview of the process.
from deeprank.learn import *
from deeprank.learn.model3d import cnn_reg
import torch.optim as optim
import numpy as np
# preprocessed input data
hdf5_path = 'train_data.hdf5'
# output directory
out = './my_deeplearning_train/'
# declare the dataset instance
data_set = DataSet(
hdf5_path,
grid_info={
'number_of_points': (10, 10, 10),
'resolution': (10, 10, 10)
},
select_feature='all',
select_target='class',
)
# create the network
model = NeuralNet(data_set,cnn_reg,model_type='3d',task='reg',
cuda=False)
# change the optimizer (optional)
model.optimizer = optim.SGD(model.net.parameters(),
lr=0.001,
momentum=0.9,
weight_decay=0.005)
# start the training, this will generate a model file named `best_valid_model.pth.tar`.
model.train(nepoch=50, divide_trainset=0.8, train_batch_size=5, num_workers=0)In the first part of the script we create a Torch database from the HDF5 file. We can specify one or several HDF5 files and even select some conformations using the dict_filter argument. Other options of DataSet can be used to specify the features/targets the normalization, etc ...
We then create a NeuralNet instance that takes the dataset as input argument. Several options are available to specify the task to do, the GPU use, etc ... We then have simply to train the model. Simple !
Prediction is almost similar to training, apart from the fact that there are no class labels.
from deeprank.learn import *
from deeprank.learn.model3d import cnn_reg
from deeprank.models.metrics import OutputExporter
import torch.optim as optim
import numpy as np
# preprocessed input data
hdf5_path = 'unseen_data.hdf5'
# output directory
out = './my_deeplearning_train/'
# declare the dataset instance
data_set = DataSet(
hdf5_path,
grid_info={
'number_of_points': (10, 10, 10),
'resolution': (10, 10, 10)
},
select_feature='all',
select_target='class',
)
# create the network
model = NeuralNet(data_set, cnn_reg, model_type='3d', task='reg',
pretrained_model="best_valid_model.pth.tar",
metrics_exporters=[OutputExporter(out)],
cuda=False)
# do the prediction
model.test()If you have questions or find a bug, please report the issue in the Github issue channel.
If you want to change or further develop DeepRank-mut code, please check the Developer Guideline to see how to conduct further development.