Request code to implement XENet paper · pyg-team/pytorch_geometric#8257

2023-10-24T08:58:16.000Z

### 🚀 The feature, motivation and pitch - Paper: XENet: Using a new graph convolution to accelerate the timeline for protein design on quantum computers - This paper introduces XENet, a new graph convolutional network designed to improve the representation of protein environments in sequence design protocols. By paying attention to both incoming and outgoing edge attributes, XENet is able to better model local kinematics problems such as protein design. The paper compares XENet to existing graph convolutions and demonstrates its ability to decrease rotamer sample counts in Rosetta's rotamer substitution protocol by 40% without loss in quality. This allows larger protein design problems to fit onto near-term quantum computers. Additionally, XENet displays an ability to handle deeper architectures than competing convolutions. The authors believe that their work is relevant to the field of protein design and can lead to further advancements in artificial intelligence brought forth by graph neural networks. - Main feature: XEnet can take node features and edge features as input, output potential node features and edge features, and can be used for both node classification and edge classification. ### Alternatives The code solution is: In TensorFlow 2, We can use the XENetConv class in the Spektral library to implement the Xception convolutional layer. ### Additional context ![image](https://github.com/pyg-team/pytorch_geometric/assets/100210376/c21f27d9-7e51-4ecb-92ba-4321fd529ca3)

(7 留言) (1 反應) (0 負責人)Python (19,985 star) (3,514 fork)batch import

featuregood first issuehelp wanted

描述

🚀 The feature, motivation and pitch

Paper: XENet: Using a new graph convolution to accelerate the timeline for protein design on quantum computers
This paper introduces XENet, a new graph convolutional network designed to improve the representation of protein environments in sequence design protocols. By paying attention to both incoming and outgoing edge attributes, XENet is able to better model local kinematics problems such as protein design. The paper compares XENet to existing graph convolutions and demonstrates its ability to decrease rotamer sample counts in Rosetta's rotamer substitution protocol by 40% without loss in quality. This allows larger protein design problems to fit onto near-term quantum computers. Additionally, XENet displays an ability to handle deeper architectures than competing convolutions. The authors believe that their work is relevant to the field of protein design and can lead to further advancements in artificial intelligence brought forth by graph neural networks.
Main feature: XEnet can take node features and edge features as input, output potential node features and edge features, and can be used for both node classification and edge classification.

Alternatives

The code solution is: In TensorFlow 2, We can use the XENetConv class in the Spektral library to implement the Xception convolutional layer.

Additional context

貢獻者指南

技術棧: python
領域: machine learning
議題類型: feature
難度: 5
預計時間: over 1 week
活動狀態: fresh
清晰度: needs investigation
前置要求: Graph neural networksPyTorchPaper implementation experience
新手友善度: 15
研究方向: Review the XENet paper (linked in the issue) to understand the proposed graph convolution. Examine existing convolution implementations in PyTorch Geometric (e.g., GCNConv, GATConv) for structure. Check the issue comments for any maintainer guidance or clarification on the desired implementation details. Note that the issue references Spektral's XENetConv as an alternative; consider how to adapt the Keras based design to PyTorch Geometric's API.