Predicting Drug-Gene Relations via Analogy Tasks with Word Embeddings

2024-06-03

arXiv.org

AI Summary

This study addresses drug–target relationship prediction by proposing a lightweight analogical reasoning method grounded in biomedical word embeddings. Methodologically, it pioneers the direct application of vector-space analogies (e.g., “drug A : gene A :: drug B : ?”) to cross-entity relational inference, integrating BioConceptVec pretrained embeddings with self-supervised PubMed abstract word vectors, and incorporating KEGG/Reactome pathway semantics to enhance biological plausibility. Its key contributions are: (1) replacing complex graph neural networks or large language models with interpretable, embedding-based arithmetic operations; and (2) demonstrating strong temporal generalization—accurately predicting newly discovered drug targets using only historically known relationships. On standard benchmarks, the method achieves accuracy comparable to GPT-4, validating the efficacy and practicality of simple analogical reasoning for biomedical knowledge discovery.

Technology Category

• Cognitive Modeling & Cognitive Systems: Analogy
• Reasoning under Uncertainty: Relational Probabilistic Models
• Data Mining & Knowledge Management: Semantic Web

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Natural language processing (NLP) is utilized in a wide range of fields, where words in text are typically transformed into feature vectors called embeddings. BioConceptVec is a specific example of embeddings tailored for biology, trained on approximately 30 million PubMed abstracts using models such as skip-gram. Generally, word embeddings are known to solve analogy tasks through simple vector arithmetic. For instance, $mathrm{ extit{king}} - mathrm{ extit{man}} + mathrm{ extit{woman}}$ predicts $mathrm{ extit{queen}}$. In this study, we demonstrate that BioConceptVec embeddings, along with our own embeddings trained on PubMed abstracts, contain information about drug-gene relations and can predict target genes from a given drug through analogy computations. We also show that categorizing drugs and genes using biological pathways improves performance. Furthermore, we illustrate that vectors derived from known relations in the past can predict unknown future relations in datasets divided by year. Despite the simplicity of implementing analogy tasks as vector additions, our approach demonstrated performance comparable to that of large language models such as GPT-4 in predicting drug-gene relations.

Problem

• Predict drug-gene relations using analogy tasks with embeddings
• Improve prediction accuracy by categorizing drugs and genes via pathways
• Compare performance of embedding-based approach with large language models

Innovation

• Uses BioConceptVec embeddings for biology
• Predicts drug-gene relations via analogy tasks
• Improves performance with biological pathway categorization