CoFormerSurv: Collaborative transformer for multi-omics survival analysis

Area of Science:

• Biomedicine
• Computational Biology
• Genomics

Background:

• High-throughput sequencing generates vast multi-omics data, crucial for understanding disease complexity.
• Multi-omics survival analysis improves prognostic predictions for personalized medicine.
• Transformer architectures show promise but face challenges with right-censored data in multi-omics survival analysis.

Purpose of the Study:

• To propose an innovative collaborative Transformer framework, CoFormerSurv, for multi-omics survival analysis.
• To effectively utilize Transformer architecture for extracting complementary information across different omics.
• To enhance survival prediction performance by addressing challenges in modeling right-censored data.

Main Methods:

• Developed CoFormerSurv, a collaborative Transformer framework with two architectures: inter-omics Transformer and inter-sample graph Transformer.
• The inter-omics Transformer uses multi-head self-attention to capture complementary information across omics.
• The inter-sample graph Transformer integrates structural information from fused multi-omics graphs to explore sample relationships.

Main Results:

• CoFormerSurv collaboratively generates comprehensive multi-omics features.
• The framework improves Cox-PH model performance in survival analysis.
• Experimental results demonstrate superior performance over single-Transformer architectures and existing models.

Conclusions:

• CoFormerSurv effectively explores complementary information from both inter-omics and cross-sample perspectives.
• The proposed method advances multi-omics survival analysis for improved prognostic predictions.
• This framework holds potential for developing more effective personalized treatment strategies in precision medicine.