Slimformer: An NLP-based web server for semantic categorization of gene sets.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Omics data analysis generates large gene sets, complicating interpretation of cellular mechanisms.
• Current categorization methods often overlook semantic similarities in textual descriptions, relying heavily on hierarchical ontologies like Gene Ontology.
• There is a need for advanced methods to integrate linguistic and functional information for comprehensive gene set analysis.

Purpose of the Study:

• To develop and validate Slimformer, an embedding-based Natural Language Processing model for enhanced gene set categorization.
• To leverage contextual relationships from gene set names, descriptions, and associated genes for improved classification.
• To provide a flexible framework for systematic gene set categorization, aiding in the interpretation of omics data.

Main Methods:

• Developed Slimformer, an embedding-based Natural Language Processing model.
• Trained a supervised classifier on a manually curated gold standard dataset of annotated gene sets.
• Utilized gene set names, descriptions, and associated genes to learn contextual relationships.
• Applied the model to 2856 annotated gene sets and to gene expression data from RSV-infected human cells.

Main Results:

• Slimformer achieved 82.4% balanced accuracy and an F1-score of 0.867 on annotated gene sets.
• The model identified significant downregulation of cell cycle processes in RSV-infected cells, a finding missed by other tools.
• Demonstrated improved interpretability of omics data through integrated linguistic and functional information.

Conclusions:

• Slimformer offers a novel and effective approach to gene set categorization by incorporating semantic similarity.
• The tool enhances the interpretability of omics data, facilitating the discovery of biological insights.
• Slimformer provides a valuable framework for researchers analyzing complex biological datasets and understanding disease mechanisms.