BTR-MCL: A Bioinformatics Tool Recommendation Method With Multi-View Contrastive Learning

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• The rapid growth of digital information in bioinformatics has led to numerous tools.
• Existing methods struggle to effectively manage and recommend these bioinformatics tools.
• Current recommendation systems lack the ability to leverage descriptive information and correlations for sparse interaction data.

Purpose of the Study:

• To develop an effective recommendation system for bioinformatics tools.
• To overcome the limitations of existing recommendation methods in handling tool-centric data and sparse interactions.
• To enhance the discovery and utilization of bioinformatics resources.

Main Methods:

• Collected quantitative bioinformatics operations, tools, and contextual data from https://bio.tools.
• Constructed a knowledge graph dataset for bioinformatics tools.
• Proposed the Bioinformatics Tool Recommendation Method with Multi-view Contrastive Learning (BTR-MCL).
• Utilized BERT for entity semantic enrichment and implemented multi-view contrastive learning (TT view, OT view, KG view).

Main Results:

• BTR-MCL significantly improves upon state-of-the-art baselines in top-K recommendation tasks.
• Achieved a Recall@10 metric improvement of over 82.71% compared to the best baseline.
• Demonstrated enhanced tool embedding through local and global contrastive learning.

Conclusions:

• BTR-MCL offers a robust solution for bioinformatics tool recommendation.
• The multi-view contrastive learning approach effectively addresses data sparsity and item-dominated recommendation challenges.
• This method enhances the semantic understanding and recommendation performance for bioinformatics tools.