how to use the GearNet to extract the protein feature?

[onlyonewater]

hi, authors, a great work, I want to use the GearNet as a feature extractor to extract the protein features, how to use it?

thanks!!!

[Oxer11]

Hi! If you want to extract protein features with GearNet, I suggest to read this tutorial carefully. Basically, you only need to write your customized dataset and then use GearNet as your encoder.

[onlyonewater]

ok, got it, I will have a try. thanks!

[pearl-rabbit]

Hello, I haven't used pretrained models before, and after reading the documentation, I still feel a bit confused. I want to use my own PDB dataset, but I don't know how to load it and apply the pretrained model to obtain its representation

[Oxer11]

Hi, to use the pre-trained models, please check the last section of this tutorial for pre-training and fine-tuning. You will find how to load the pre-trained model and fine-tune on your own dataset.

[pearl-rabbit]

Hello, I'm sorry to bother you again. I hope to use GearNet loaded with pre-trained weights to extract protein features, but the program has neither output nor termination. What's the problem?

# protein
protein = data.Protein.from_pdb(pdb_file, atom_feature="position", bond_feature="length", residue_feature="symbol")
_protein = data.Protein.pack([protein])
protein = graph_construction_model(_protein)

# model
gearnet_edge = models.GearNet(input_dim=21, hidden_dims=[512, 512, 512, 512, 512, 512],
                              num_relation=7, edge_input_dim=59, num_angle_bin=8,
                              batch_norm=True, concat_hidden=True, short_cut=True, readout="sum")
pthfile = 'models/angle_gearnet_edge.pth'
net = torch.load(pthfile)
gearnet_edge.load_state_dict(net)

#output
with torch.no_grad():
    output = gearnet_edge(protein, protein.node_feature.float(), all_loss=None, metric=None)

[Oxer11]

Hi, the code looks good to me.

Could you provide more information about the bug? What does it mean by no output from the program? It seems that you haven't included a print in your code. If the code runs without termination, could you please show which part the code will stuck at?

[pearl-rabbit]

The program has been running the last line of code：

output = gearnet_edge(protein, protein.node_feature.float(), all_loss=None, metric=None)

[Oxer11]

I think it's simply because the model hasn't finished yet, since you're running the code on CPU without putting the model on GPU.

[pearl-rabbit]

I tried to replace 'utils. sparse_coo_tensor' with 'torch. sparse_coo_tensor' in line 802 of "layers.conv.py" , and the program was able to continue executing (although an error was reported later).

[Oxer11]

This may be due to the compilation problem of torch_ext. You can check this issue. https://github.com/DeepGraphLearning/torchdrug/issues/8#issuecomment-916706055

[Oxer11]

BTW, to generate the embeddings, don't forget to switch the model to .eval() mode by calling gearnet_edge.eval().

[pearl-rabbit]

Is it written like this?

gearnet_edge.eval()
output = gearnet_edge(graph=protein, input=protein.node_feature.float())

[Oxer11]

Yes!

[jinzhuwei]

Hello, I haven't used pretrained models before, and after reading the documentation, I still feel a bit confused. I want to use my own PDB dataset, but I don't know how to load it and apply the pretrained model to obtain its representation

Hello, I want to consult about you using your own PDB dataset to build Graph, whether to implement, I also want to build a protein graph on my own dataset, I want to ask you about the implementation of this part of the sale, I hope to get your sharing, thank you very much.

[Tizzzzy]

Is it written like this?

gearnet_edge.eval()
output = gearnet_edge(graph=protein, input=protein.node_feature.float())

Hi pearl-rabbit: I am also trying to figure out how to use GearNet loaded with pre-trained weights to extract protein features. I am writing the same code as you, and below is my code:

import os
import sys
import argparse
import torch
from torchdrug import core
sys.path.append(os.path.dirname(os.path.dirname(__file__)))
from gearnet.model import GearNetIEConv
from torchdrug.data import Protein
from torchdrug.core import Registry as R
from torchdrug import data, utils
from torchdrug import layers
from torchdrug.layers import geometry
from torchdrug import models

pdb_file = utils.download("https://files.rcsb.org/download/2LWZ.pdb", "./")

graph_construction_model = layers.GraphConstruction(node_layers=[geometry.AlphaCarbonNode()], 
                                                    edge_layers=[geometry.SpatialEdge(radius=10.0, min_distance=5),
                                                                 geometry.KNNEdge(k=10, min_distance=5),
                                                                 geometry.SequentialEdge(max_distance=2)],
                                                    edge_feature="gearnet")

# protein
protein = Protein.from_pdb(pdb_file, atom_feature="position", bond_feature="length", residue_feature="symbol")
_protein = Protein.pack([protein])
protein = graph_construction_model(_protein)

# model
gearnet_edge = models.GearNet(input_dim=21, hidden_dims=[512, 512, 512, 512, 512, 512],
                              num_relation=7, edge_input_dim=59, num_angle_bin=8,
                              batch_norm=True, concat_hidden=True, short_cut=True, readout="sum")
pthfile = '/content/angle_gearnet_edge.pth'
net = torch.load(pthfile, map_location=torch.device('cpu'))
gearnet_edge.load_state_dict(net)

#output
with torch.no_grad():
    gearnet_edge.eval()
    print(protein)
    output = gearnet_edge(protein, protein.node_feature.float(), all_loss=None, metric=None)
    print(output)

However, I am getting following error:

Traceback (most recent call last):
  File "/content/GearNet-main/script/test.py", line 40, in <module>
    output = gearnet_edge(protein, protein.node_feature.float(), all_loss=None, metric=None)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torchdrug/models/gearnet.py", line 95, in forward
    hidden = self.layers[i](graph, layer_input)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torchdrug/layers/conv.py", line 91, in forward
    update = self.message_and_aggregate(graph, input)
  File "/usr/local/lib/python3.10/dist-packages/torchdrug/layers/conv.py", line 813, in message_and_aggregate
    return update.view(graph.num_node, self.num_relation * self.input_dim)
RuntimeError: shape '[57, 147]' is invalid for input of size 1197

Can you please take a look at my code? Thank you so much