liupf/KAR4DDI

Overview

The dataset in this study is a Drug-Drug Interaction Event (DDIE) dataset obtained from DeepDDI 2. It includes detailed information on 2,386 drugs, each represented by a 50-dimensional Principal Components Analysis (PCA) feature vector, and the corresponding SMILES strings. Additionally, drug descriptions from DDInter and DrugBank have been integrated into the dataset.

The DDIE dataset comprises 222,127 drug pairs, enabling the prediction of 113 different DDIE types. Addressing few-shot scenarios is crucial due to the frequent occurrence of rare and poorly documented drug interactions in clinical settings, which present significant challenges.

Data Sample Distribution

The following table shows the distribution of data samples across different interaction frequency categories in the training, validation, and test sets:

Freq	Train	Valid	Test
Common	44,126	44,113	132,110
Few	108	128	298
Rare	43	34	85

This categorization into 'common', 'few', and 'rare' is based on the frequency of DDIE occurrences, which helps address the challenges posed by different frequency categories in real-world scenarios. Additionally, categories with fewer than two samples were removed, and the remaining samples were distributed into training, validation, and test sets at ratios of 2:2:6 to enhance the dataset's quality and reliability.

Knowledge Extraction

In our approach to knowledge extraction, we transform drug features into drug types. These features are then subjected to dimensionality reduction via t-SNE, reducing them to a 2-dimensional space suitable for visualization and clustering. This method is preferred for its effectiveness in preserving the local structure of the data while reducing dimensionality, which facilitates the identification of patterns and relationships inherent in the high-dimensional space.

For clustering the 2-dimensional drug features, we employ several algorithms that require a predefined number of clusters, such as K-means, Birch, and Agglomerative clustering. These techniques are chosen for their appropriateness for quantitative analysis and compatibility with dimensionality reduction techniques like t-SNE.