xzenglab / tagppi Goto Github PK

老师，您好，最近在研究生物信息学领域读到了您的论文，深受启发。我想要进行一些实验加深对文章的理解，但是在编译的时候遇到了如下问题：
dgl._ffi.base.DGLError: [12:07:26] /opt/dgl/src/array/array_arith.cc:81: Operator Add does not support cuda device.
代码定位在如下位置：
File "/TAGPPI-main/my_train_and_validation.py", line 57, in train
y_pred = attention_model(dgl.batch(G1),pad_dmap(dmap1),dgl.batch(G2), pad_dmap(dmap2))
我根据它定位的语句修改了attention_model这个模型对应的forward函数里的加法操作。修改的语句如下：
gc1=torch.add((1-w1)g1,w1seq1)和 gc2=torch.add((1-w1)g2,w1seq2)
修改后的语句为：
y1=(1-w1)g1
y2=w1seq1
gc1 = torch.add(y1.to('cpu'),y2.to('cpu'))
x1=(1-w1)g2
x2=w1seq2
gc2=torch.add(x1.to('cpu'),x2.to('cpu')
再次运行代码还是会出现如下问题：
dgl._ffi.base.DGLError: [12:07:26] /opt/dgl/src/array/array_arith.cc:81: Operator Add does not support cuda device.
我不知道它是因为什么原因引起的，不知您是否可以解答我的困惑。谢谢！

Why?

Hello.

I am very interested in TAGPPI, so I am reproducing the paper and would like to compare it with other methods. But no matter how I adjust the model parameters and learning rate, after multiple epochs of training, the loss in the training set always remains around 0.69 and the acc remains around 0.5. At the same time, I also checked the embedding obtained by SeqVec and did not find any errors.May I ask where I may have made an error?
Thank you for your help!

作者，您好，感谢您开源代码，在文中的Construction of predicted contact map这一节您说明了contact map的生成规则，请问您可以告诉我生成contact map的具体代码吗？我在https://alphafold.ebi.ac.uk/download下载了数据，但是解压之后是pdb文件，请问我是需要对这个数据进行处理然后生成您提供的数据中npz文件吗？

如果不能告知处理细节，请问您可以告诉我能参考什么代码吗？

十分感谢～打扰了

Hello,
I am interested in running TAGPPI prediction on a list of possible PPIs to compare to a results I obtained using a different machine learning tool. Are there trained-model files available that can be used to run such a "quick check" analysis without training a new model myself? The multi-species dataset model presented in the publication would be the most interesting for me, but could make due with the yeast model. (I've downloaded these datasets, but having an already trained and cross-validated model would be really great).

From there, as I understand, I would to use SeqVec separately to generate embeddings for my test data (or is this integrated in the TAGPPI pipeline?). Then, could you please outline which .py scripts I would need to run and which variables need to be set in them to make predictions on my test PPI pairs using the multi-species trained model? (I think all I will need in the end is the 1. trained model file, 2. protein embeddings, 3. a pairwise tsv of each PPI I wish to test, with names matching the input fasta to SeqVec?) I see some comments in the py files, but how exactly to use the tool isn't entirely clear to me.

Thank you for the help!

Hello, while troubleshooting TagPPI to work on our systems, I've tried recreating the multi-species (40% filter) model with the data made available.
(I used Seqvec to embed CeleganDro...dictionary.tsv, downloaded your mul_cmap alphafold data, and am training/testing on the multi-40.tsv's)

While training, torch gives this error:

Traceback (most recent call last):
  File "/lustre/fs0/home/iwill/TAGPPI/TAGPPI-main/my_main.py", line 25, in <module>
    main()
  File "/lustre/fs0/home/iwill/TAGPPI/TAGPPI-main/my_main.py", line 22, in main
    train(trainArgs)
  File "/lustre/fs0/home/iwill/TAGPPI/TAGPPI-main/my_train_and_validation.py", line 56, in train
    y_pred = attention_model(dgl.batch(G1),pad_dmap(dmap1),dgl.batch(G2), pad_dmap(dmap2))
  File "/home/iwill/my-envs/tagppi_6/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/lustre/fs0/home/iwill/TAGPPI/TAGPPI-main/TAGlayer.py", line 78, in forward
    seq1 = self.textcnn(pad_dmap1)
  File "/home/iwill/my-envs/tagppi_6/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/lustre/fs0/home/iwill/TAGPPI/TAGPPI-main/TAGlayer.py", line 29, in forward
    features = self.mx3(features)
  File "/home/iwill/my-envs/tagppi_6/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/iwill/my-envs/tagppi_6/lib/python3.10/site-packages/torch/nn/modules/pooling.py", line 92, in forward
    return F.max_pool1d(input, self.kernel_size, self.stride,
  File "/home/iwill/my-envs/tagppi_6/lib/python3.10/site-packages/torch/_jit_internal.py", line 484, in fn
    return if_false(*args, **kwargs)
  File "/home/iwill/my-envs/tagppi_6/lib/python3.10/site-packages/torch/nn/functional.py", line 696, in _max_pool1d
    return torch.max_pool1d(input, kernel_size, stride, padding, dilation, ceil_mode)
RuntimeError: max_pool1d() Invalid computed output size: -21

Using an older version gives the same result, but perhaps with a little more detail in the error message:

  File "/home/iwill/my-envs/tagppi_5/lib/python3.6/site-packages/torch/nn/functional.py", line 653, in _max_pool1d
    return torch.max_pool1d(input, kernel_size, stride, padding, dilation, ceil_mode)
RuntimeError: Given input size: (128x1x108). Calculated output size: (128x1x-21). Output size is too small

Thoughts on getting this to run?
Thank you