Boost-RS

Boost-RS enhances prediction of enzyme-substrate interactions by boosting embedding vectors in collaborative filtering recommender systems through integration of heterogeneous auxiliary relational data and contrastive learning.

Key Features:

• Embedding boosting: Boosts embedding vectors by integrating heterogeneous auxiliary data, including relational data such as hierarchical structures, pairwise relationships, and groupings.
• Auxiliary learning tasks: Trains and dynamically tunes models across multiple relevant auxiliary learning tasks to improve embedding quality.
• Contrastive learning: Employs contrastive learning tasks to exploit relational auxiliary data and enhance discrimination in embeddings.
• Collaborative filtering integration: Enhances collaborative filtering (CF) recommender systems for enzyme-substrate interaction prediction.
• Comparative methods: Demonstrates improvements over attribute concatenation and multi-label learning approaches.
• Evaluation and analysis: Uses ablation studies and visualization techniques to assess the contribution of each auxiliary task to embedding learning.
• Data source: Applied to enzyme-substrate interaction prediction using interaction data from the KEGG database.

Scientific Applications:

• Enzyme promiscuity characterization: Facilitates exploration of unexplored and under-documented enzyme promiscuity on substrates.
• Pathway design: Supports construction of novel biomolecule synthesis pathways by improving enzyme-substrate interaction predictions.
• Metabolite identification: Aids identification of metabolic products from ingested compounds via improved interaction inference.
• Xenobiotic metabolism: Assists elucidation of xenobiotic metabolism through enhanced prediction of enzyme-substrate relationships.

Methodology:

Integrates heterogeneous auxiliary relational data to boost embedding vectors; trains and dynamically tunes across multiple auxiliary learning tasks; employs contrastive learning tasks to exploit relational data; applies the approach to baseline collaborative filtering models and evaluates contributions via ablation studies and visualization.