Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Knowledge-augmented Graph Neural Networks with Concept-aware Attention for Adverse Drug Event Detection

Shaoxiong Ji, Ya Gao, Pekka Marttinen

TL;DR

This paper addresses adverse drug event (ADE) detection by integrating explicit medical knowledge into a graph-based model. It introduces KnowCAGE, a Knowledge-augmented Concept-Aware Graph Embedding framework that constructs a heterogeneous text graph (documents, words, concepts) augmented with UMLS concepts and employs a concept-aware attention mechanism to learn node-type-specific features, followed by ensemble classification with contextual embeddings. The approach yields competitive or superior results across four public ADE datasets, with the concept-aware attention consistently outperforming other attention schemes and knowledge augmentation providing notable gains over baselines. The method enhances both predictive accuracy and interpretability, offering practical benefits for biomedical text mining and pharmacovigilance applications.

Abstract

Adverse drug events (ADEs) are an important aspect of drug safety. Various texts such as biomedical literature, drug reviews, and user posts on social media and medical forums contain a wealth of information about ADEs. Recent studies have applied word embedding and deep learning -based natural language processing to automate ADE detection from text. However, they did not explore incorporating explicit medical knowledge about drugs and adverse reactions or the corresponding feature learning. This paper adopts the heterogenous text graph which describes relationships between documents, words and concepts, augments it with medical knowledge from the Unified Medical Language System, and proposes a concept-aware attention mechanism which learns features differently for the different types of nodes in the graph. We further utilize contextualized embeddings from pretrained language models and convolutional graph neural networks for effective feature representation and relational learning. Experiments on four public datasets show that our model achieves performance competitive to the recent advances and the concept-aware attention consistently outperforms other attention mechanisms.

Knowledge-augmented Graph Neural Networks with Concept-aware Attention for Adverse Drug Event Detection

TL;DR

This paper addresses adverse drug event (ADE) detection by integrating explicit medical knowledge into a graph-based model. It introduces KnowCAGE, a Knowledge-augmented Concept-Aware Graph Embedding framework that constructs a heterogeneous text graph (documents, words, concepts) augmented with UMLS concepts and employs a concept-aware attention mechanism to learn node-type-specific features, followed by ensemble classification with contextual embeddings. The approach yields competitive or superior results across four public ADE datasets, with the concept-aware attention consistently outperforming other attention schemes and knowledge augmentation providing notable gains over baselines. The method enhances both predictive accuracy and interpretability, offering practical benefits for biomedical text mining and pharmacovigilance applications.

Abstract

Adverse drug events (ADEs) are an important aspect of drug safety. Various texts such as biomedical literature, drug reviews, and user posts on social media and medical forums contain a wealth of information about ADEs. Recent studies have applied word embedding and deep learning -based natural language processing to automate ADE detection from text. However, they did not explore incorporating explicit medical knowledge about drugs and adverse reactions or the corresponding feature learning. This paper adopts the heterogenous text graph which describes relationships between documents, words and concepts, augments it with medical knowledge from the Unified Medical Language System, and proposes a concept-aware attention mechanism which learns features differently for the different types of nodes in the graph. We further utilize contextualized embeddings from pretrained language models and convolutional graph neural networks for effective feature representation and relational learning. Experiments on four public datasets show that our model achieves performance competitive to the recent advances and the concept-aware attention consistently outperforms other attention mechanisms.
Paper Structure (19 sections, 7 equations, 4 figures, 6 tables)

This paper contains 19 sections, 7 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Knowledge-augmented Graph Construction
  5. Heterogeneous Graph Convolution
  6. Concept-aware Attention Mechanism
  7. Classification Layers and Model Training
  8. Experimental Setup
  9. Data and Preprocessing
  10. Baselines and Evaluation
  11. Hyper-parameters
  12. Results
  13. Comparison with Baselines in Different Domains (RQ1)
  14. Usefulness of the Knowledge Augmented Graph Convolution (RQ2)
  15. Effectiveness of the Concept-Aware Attention (RQ3)
  16. ...and 4 more sections

Figures (4)

  • Figure 1: An example of a text mentioning an adverse drug event from karimi2015cadec. The recognition of drugs and adverse reactions requires medical knowledge and relational reasoning.
  • Figure 2: An illustration of the model architecture with knowledge-augmented graph embeddings and concept-aware representations
  • Figure 3: The effect of different computational methods for weights of edges.
  • Figure 4: A visualization of the concept-aware attention in the form of node cloud