peizhenbai / drugban Goto Github PK

Dear Authors,

How long does this method need to train?

你好，我想问下你这个训练拟合，一共训练了多久呀？我看代码里面好像是只有一层gcn？

您好，想向您请教一下为何药物分子的最大原子数设置为290？

Can you share the code for your clustering-based pair split?

在您的

why use test_target_dataset as both val_generator and test generator? it will lead to data leakage.

Hello, I am very interested in your research. Do you have an example code of "Interoperability with linear attention visualization"？

您好，
我一直在研究您的项目，并且对其中的某些部分有一些疑问，特别是如何通过Attention Map映射到对蛋白结合有贡献的原子。
具体来说，我注意到您在预处理中添加了虚拟节点，但我不太清楚它们是如何与Attention Map相关联的。是否有特定的方式或算法，您可以向我解释一下吗？
我非常感激您的时间和帮助，这将有助于我更深入地理解您的项目。
谢谢！

你好~，我想问一下这个DrugBAN是否是一个已经训练好的模型（或者说有训练好的版本），我想要用我自己的蛋白质和分子数据能否直接进行预测是否可行，还是说我应该用一些已知结合情况的数据进行训练才可以使用呢？

Dear Doctor Bai:
The work DrugBAN is definitely excellent work! For it combines the domain adversarial network and interpretability, both essential problems in drug discovery.
However, I can not understand completely when the target domain data are predicted if the interpretability module will work in the target domain Proteins and Drugs.
For the output of attention module may be influenced by the CDAN.
(I do not know much of it maybe because of my reading of the preprint version of your paper neither the version in NMI).
Thanks for your reading my question again!

Hi:

It is really a nice work about DTI.
I'm a little confused by the human dataset in this github repo. datasets/human/full.csv has 6728 samples. However, the sum of samples in train.csv, val.csv and test.csv in datasets/human/random/ is 5997.

So are there some special filters for human dataset?

Thanks