PCLSurv: a prototypical contrastive learning-based multi-omics data integration model for cancer survival prediction

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Accurate cancer survival prediction is crucial but challenging due to complex multi-omics data.
• Existing methods often fail to capture comprehensive features for precise predictions.

Purpose of the Study:

• To introduce PCLSurv, a deep learning framework for enhanced cancer survival prediction using multi-omics data.
• To improve patient stratification by effectively integrating diverse molecular and clinical information.

Main Methods:

• PCLSurv utilizes autoencoders for omics-specific feature extraction.
• Sample-level and prototypical contrastive learning are employed for feature representation and alignment.
• A bilinear fusion module integrates extracted features into a unified representation.

Main Results:

• PCLSurv demonstrated superior performance in cancer survival prediction across 11 cancer datasets.
• The framework effectively distinguishes patient groups with varying survival outcomes.
• Experimental results confirm PCLSurv's advantage over existing prediction methods.

Conclusions:

• PCLSurv offers a robust framework for cancer survival prediction and patient stratification.
• The integration of multi-omics data through advanced deep learning techniques significantly enhances prediction accuracy.
• The developed model provides a valuable tool for clinical oncology decision-making.