Learning temporal difference embeddings for biomedical hypothesis generation.

Area of Science:

• Biomedical informatics
• Computational biology
• Data science

Background:

• Hypothesis generation (HG) is crucial for drug discovery, predicting side effects, and precision treatment.
• Existing HG methods often fail to capture the dynamic evolution of scientific term relationships.

Purpose of the Study:

• To propose a novel temporal difference embedding (TDE) learning framework for hypothesis generation.
• To model the temporal evolution of term-pair relations for predicting future interactions.

Main Methods:

• Formulating HG as a future connectivity prediction task on a dynamic attributed graph.
• Modeling local neighbor changes and global graph structure changes over time.
• Learning local and global temporal difference embeddings (TDE) for node-pairs.
• Inferring future term-pair relations using a recurrent network.

Main Results:

• The TDE framework effectively models the temporal difference information evolution of term-pair relations.
• Experiments on three real-world biomedical datasets demonstrate the approach's effectiveness and superiority.
• The method successfully predicts future interactions between scientific terms.

Conclusions:

• The proposed TDE learning framework offers a significant advancement in hypothesis generation.
• Accurate modeling of temporal dynamics in scientific term relations is key for improved predictions.
• The approach has strong implications for accelerating drug discovery and precision medicine.