MRGNN

class MRGNN(*, molecule_channels=66, hidden_channels=32, middle_channels=16, layer_count=4, out_channels=1)[source]

Bases: chemicalx.models.base.Model

An implementation of the MR-GNN model from [xu2019].

See also

This model was suggested in https://github.com/AstraZeneca/chemicalx/issues/12

xu2019

Xu, N., et al. (2019). MR-GNN: Multi-resolution and dual graph neural network for predicting structured entity interactions. IJCAI International Joint Conference on Artificial Intelligence, 2019, 3968–3974.

Methods Summary

forward(molecules_left, molecules_right)

Run a forward pass of the MR-GNN model.

unpack(batch)

Return the left molecular graph and right molecular graph.

Methods Documentation

forward(molecules_left, molecules_right)[source]

Run a forward pass of the MR-GNN model.

Parameters
  • molecules_left (PackedGraph) – Batched molecules for the left side drugs.

  • molecules_right (PackedGraph) – Batched molecules for the right side drugs.

Return type

FloatTensor

Returns

A column vector of predicted synergy scores.

unpack(batch)[source]

Return the left molecular graph and right molecular graph.