Zuobai Zhang

Results 52 comments of Zuobai Zhang

Sorry for the inconvience! This is because I set the default files as v2 datasets instead of v4 datasets. The easiest way to change this is to inherit the `datasets.AlphaFoldDB`...

Hi, thanks for the question. In the original experiment, we train all models for 200 epochs. When I organize the code, I find training for 50 epochs can yield similar...

Actually, I think the downstream performance is very sensitive to batch size. For the current config, I find `batch_size=8` is the best setting. Typically, if you want to change the...

Hi, I'm wondering how you use the `graph_construction_model`. Usually, we use `graph_contruction_model` to define a graph with 7 edge types and perform relational message passing. The implementation of relational message...

Hi, the config file for GearNet-Edge-IEConv on Fold is [`config/Fold3D/gearnet_edge_ieconv.yaml`](https://github.com/DeepGraphLearning/GearNet/blob/main/config/downstream/Fold3D/gearnet_edge_ieconv.yaml). The pre-trained checkpoints of GearNet-Edge can be found at https://zenodo.org/record/7723075.

Hi! I think the error is that you forget to set `protein_.view = 'residue'`. This is done by using the `transforms.ProteinView` in the tutorial. If you don't do this, the...

I see. The original pre-trained checkpoints were deleted by my cluster. I've pre-trained a new GearNet-Edge-IEConv recently. You can download the checkpoint from [this link](https://udemontreal-my.sharepoint.com/:u:/g/personal/zuobai_zhang_umontreal_ca/ERfr0L1IFi1PlH7vNRfuLxkBSGZ_KdQMoyEUIHjpxXN_tw?e=cZYSfr) and have a try. Please...

Hi, sorry for the confusion. Could you please check again whether you pass the `residue_feature` instead of `atom_feature` into the model? If this works correctly, I think the shape of...

Hi, thanks for the good words! To use GearNet on your customized dataste, you can follow [this tutorial](https://torchprotein.ai/tutorial_3), which describes how to use TorchDrug to construct `data.Protein` object and pass...

The situation you says is quite uncommon. I assume that the alpha carbon is missing due to some data loss issues. In this case, we'd better discard the corresponding residues,...