Adaptive risk-aware sharable and individual subspace learning for cancer survival analysis with multi-modality data

Area of Science:

• Biomedical data science
• Computational biology
• Cancer research

Background:

• Biomedical multi-modality data (multi-omics) offer comprehensive cancer insights, enhancing survival models.
• Challenges include fusing modality-sharable and individual representations and exploring risk-aware characteristics for stratification.
• Existing learning-based survival models often involve complex hyper-parameter tuning, risking suboptimal solutions.

Purpose of the Study:

• To propose an adaptive risk-aware method for learning sharable and individual subspaces in cancer survival analysis.
• To address the challenges of representation fusion and risk subgroup exploration.
• To develop an efficient parameter estimation strategy for learning-based survival models.

Main Methods:

• A novel adaptive risk-aware sharable and individual subspace learning method is proposed.
• Joint learning of sharable and individual subspaces using intra-modality complementarity and inter-modality incoherence terms.
• Incorporation of a grouping co-expression constraint for risk-aware representation and local consistency.
• An adaptive-weighted strategy for efficient parameter estimation during training.

Main Results:

• The proposed method demonstrates superior performance in cancer survival analysis.
• Experimental results on three public datasets validate the model's effectiveness.
• The approach successfully learns and fuses modality-specific and shared features for improved prognosis.

Conclusions:

• The developed method effectively addresses key challenges in multi-modality cancer survival analysis.
• It provides a robust framework for risk stratification and prognosis evaluation.
• The adaptive strategy offers an efficient approach to parameter estimation, enhancing model utility.