Pathformer: a biological pathway informed transformer for disease diagnosis and prognosis using multi-omics data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data offers comprehensive gene regulation insights crucial for complex disease diagnosis, such as cancer.
• Conventional integration methods often neglect prior biological knowledge and lack interpretability.

Purpose of the Study:

• To develop an interpretable multi-omics integration method for disease diagnosis and prognosis.
• To enhance cancer diagnosis, survival prediction, and drug response prediction using biological pathway information.

Main Methods:

• Developed Pathformer, a biological pathway-informed Transformer model.
• Utilized a compacted multi-modal vector and pathway-based sparse neural network for multi-omics integration.
• Employed a criss-cross attention mechanism to capture pathway and modality crosstalk.

Main Results:

• Pathformer outperformed 18 comparable methods on multiple cancer datasets, showing significant improvements in F1 scores for survival, stage, and drug response prediction.
• Demonstrated biological interpretability through a case study on cancer prognosis, identifying key pathways and their crosstalk.
• Showcased potential clinical applications in cancer screening using liquid biopsy data (plasma and platelets).
• Revealed pathway deregulation and crosstalk in cancer patients' blood, suggesting potential therapeutic targets.

Conclusions:

• Pathformer provides an effective and interpretable approach for multi-omics data integration in cancer research.
• The method holds promise for advancing early cancer diagnosis, prognosis, and personalized treatment strategies.
• Identified novel biological insights into cancer pathways and their interactions with potential clinical relevance.