Torchdrug tutorial

TorchDrug

Tensor based graph

Chart

The picture is TorchDrug The basic data structure used in . They can be regarded as graphic processing libraries in the tensor framework （ for example NetworkX） Generalization , At the same time enjoy GPU Acceleration and automatic differentiation .

import torchdrug as td
from torchdrug import data

edge_list = [[0, 1], [1, 2], [2, 3], [3, 4], [4, 5], [5, 0]]
graph = data.Graph(edge_list, num_node=6)
graph.visualize()

molecular

A specific example of the graph in drug discovery is a molecule , It may have additional chemical constraints and characteristics . for example , We can SMILES String constructs a molecule .SMILES Is a language that uses linear strings to describe complex molecular structures .

mol = data.Molecule.from_smiles("C1=CC=CC=C1")
mol.visualize()
print(mol.node_feature.shape)
print(mol.edge_feature.shape)

Figure operation

A group of arbitrary graphs can also be processed in batch , To maximize the use of hardware . They can also be used cpu() and cuda() Method in CPU and GPU Transferred between .

smiles_list = ["CCSCCSP(=S)(OC)OC", "CCOC(=O)N",
               "N(Nc1ccccc1)c2ccccc2", "NC(=O)c1cccnc1"]
mols = data.PackedMolecule.from_smiles(smiles_list)
mols.visualize()
mols = mols.cuda()
print(mols)

It's like the original PyTorch Tensors are the same , Graph supports a wide range of indexing operations . Typical uses include application node masking 、 Edge shielding or graphic shielding . The following example shows how to select an edge that contains at least one carbon . See notes for more graphic operations .

node_in, node_out, bond_type = mols.edge_list.t()
edge_mask = (mols.atom_type[node_in] == td.CARBON) | \
            (mols.atom_type[node_out] == td.CARBON)
mols = mols.edge_mask(edge_mask)
mols.visualize()