NCI抗癌活性预测数据

数据集获取

git clone https://github.com/shiruipan/graph_datasets.git

curl -L -o repo.zip https://github.com/shiruipan/graph_datasets/archive/refs/heads/master.zip
unzip repo.zip

源站 README 摘录（使用方式）

A Repository of Benchmark Graph Datasets for Graph Classification

Introduction to Graph Classification

Recent years have witnessed an increasing number of applications involving objects with structural relationships, including chemical compounds in Bioinformatics, brain networks, image structures, and academic citation networks. For these applications, graph is a natural and powerful tool for modeling and capturing dependency relationships between objects.
Unlike conventional data, where each instance is represented in a feature-value vector format, graphs exhibit node–edge structural relationships and have no natural vector representation. This challenge has motivated many graph classification algorithms in recent years. Given a set of training graphs, each associated with a class label, graph classification aims to learn a model from the training graphs to predict the unseen graphs in future. The following picture shows the difference betweeb classification on vector data and graph data.

Dataset Summaization

This repository maintains 31 benchmark graph datasets, which are widely used for graph classification. The graph datasets consist of:

• chemical compounds
• citation networks
• social networks
• brain networks
  The chemical compound graph datasets are in “.sdf” or “.smi” format, and other graph dataset are represented as “.nel” format. All these graph datasets can be handle by frequent subgraph miner packages such as Moss [1] or other softwares. These graphs can be easily converted to other formats handled by Matlab or other softwares.
  A summarization of our graph datasets is given in Table 1.

If you used the dataset, please cite the related papers properly.

1. NCI Anti-cancer activity prediction data (NCI)

Description:

The NCI graph datasets are commonly used as the benchmark for graph classification. Each NCI dataset belongs to a bioassay task for anticancer activity prediction, where each chemical compound is represented as a graph, with atoms representing nodes and bonds as edges. A chemical compound is positive if it is active against the corresponding cancer, or negative otherwise. Table 1 summarizes the NCI graph data we download from PubChem. We have removed disconnected graphs and graphs with unexpected atoms (some gra

完整仓库：github.com/shiruipan/graph_datasets