README.md

EGNN - Pytorch

Implementation of E(n)-Equivariant Graph Neural Networks, in Pytorch. May be eventually used for Alphafold2 replication. This technique went for simple invariant features, and ended up beating all previous methods (including SE3 Transformer and Lie Conv) in both accuracy and performance. SOTA in dynamical system models, molecular activity prediction tasks, etc.

Install

$ pip install egnn-pytorch

Usage

import torch
from egnn_pytorch import EGNN

layer1 = EGNN(dim = 512)
layer2 = EGNN(dim = 512)

feats = torch.randn(1, 16, 512)
coors = torch.randn(1, 16, 3)

feats, coors = layer1(feats, coors)
feats, coors = layer2(feats, coors) # (1, 16, 512), (1, 16, 3)

With edges

import torch
from egnn_pytorch import EGNN

layer1 = EGNN(dim = 512, edge_dim = 4)
layer2 = EGNN(dim = 512, edge_dim = 4)

feats = torch.randn(1, 16, 512)
coors = torch.randn(1, 16, 3)
edges = torch.randn(1, 16, 16, 4)

feats, coors = layer1(feats, coors, edges)
feats, coors = layer2(feats, coors, edges) # (1, 16, 512), (1, 16, 3)

A full EGNN network

import torch
from egnn_pytorch.egnn_pytorch import EGNN_Network

net = EGNN_Network(
    num_tokens = 21,
    dim = 32,
    depth = 3,
    num_nearest_neighbors = 8
)

feats = torch.randint(0, 21, (1, 1024)) # (1, 1024)
coors = torch.randn(1, 1024, 3)         # (1, 1024, 3)
mask = torch.ones_like(feats).bool()    # (1, 1024)

feats_out, coors_out = net(feats, coors, mask = mask) # (1, 1024, 32), (1, 1024, 3)

Only attend to sparse neighbors, given to the network as an adjacency matrix. num_nearest_neighbors will need to be set to the maximum number of neighbors to cover all the neighbors as specified in the matrix.

import torch
from egnn_pytorch.egnn_pytorch import EGNN_Network

net = EGNN_Network(
    num_tokens = 21,
    dim = 32,
    depth = 3,
    num_nearest_neighbors = 3,
    only_sparse_neighbors = True
)

feats = torch.randint(0, 21, (1, 1024))
coors = torch.randn(1, 1024, 3)
mask = torch.ones_like(feats).bool()

# naive adjacency matrix
# assuming the sequence is connected as a chain, with at most 2 neighbors - (1024, 1024)
i = torch.arange(1024)
adj_mat = (i[:, None] >= (i[None, :] - 1)) & (i[:, None] <= (i[None, :] + 1))

feats_out, coors_out = net(feats, coors, mask = mask, adj_mat = adj_mat) # (1, 1024, 32), (1, 1024, 3)

Examples

To run the protein backbone denoising example, first install sidechainnet

$ pip install sidechainnet

Then

$ python denoise_sparse.py

Citations

@misc{satorras2021en,
    title 	= {E(n) Equivariant Graph Neural Networks}, 
    author 	= {Victor Garcia Satorras and Emiel Hoogeboom and Max Welling},
    year 	= {2021},
    eprint 	= {2102.09844},
    archivePrefix = {arXiv},
    primaryClass = {cs.LG}
}