SaintGSE: Transformer-based efficient and explainable gene set enrichment analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene set enrichment analysis (GSEA) is crucial for interpreting genome-wide expression data.
• Current pathway inference methods face limitations in data scarcity, model compatibility, and interpretability.
• Discovering therapeutic targets for complex diseases like osteoarthritis requires efficient and explainable models.

Purpose of the Study:

• To develop SaintGSE, an efficient and explainable model for predicting gene-pathway associations.
• To leverage a large curated differentially expressed gene (DEG) signature compendium for model training.
• To support therapeutic target discovery in complex diseases, including osteoarthritis.

Main Methods:

• Developed SaintGSE, a supervised prediction framework using an autoencoder and the SAINT transformer model.
• Trained the model on a public DEG compendium with pathway labels from EnrichR.
• Utilized Integrated Gradients (IG) for model interpretability and driver gene prioritization.
• Conducted experimental validation in mouse and human chondrocyte cultures and in vivo OA models.

Main Results:

• SaintGSE identified key OA-related pathways (NF-κB, p38) potentially modulated by Senna obtusifolia extract.
• In vitro, the extract reduced OA catabolic factors (Mmp3, Mmp13, Cox2).
• In vivo, the extract significantly reduced cartilage damage in OA mouse models.
• IG-based driver genes provided condition-specific candidates for pathway predictions.

Conclusions:

• SaintGSE provides scalable and explainable pathway prediction from DEG signatures.
• The model facilitates mechanism- and target-oriented follow-up studies.
• SaintGSE is valuable for disease research and drug candidate screening.