Generative Incomplete Multi-View Prognosis Predictor for Breast Cancer: GIMPP

Area of Science:

• Bioinformatics
• Computational Biology
• Medical Informatics

Background:

• Multi-view prediction models are crucial in healthcare and bioinformatics.
• Real-world datasets often suffer from missing views, hindering model performance.
• Incomplete multi-view prediction addresses these data challenges.

Purpose of the Study:

• To develop a generative model, GIMPP, for incomplete multi-view prediction.
• To address the challenge of missing data in breast cancer prognosis.
• To explicitly generate missing data for improved predictive accuracy.

Main Methods:

• A two-stage generative model (GIMPP) was developed.
• Stage 1: Multi-view encoder networks and bi-modal attention for shared latent representations.
• Stage 2: View-specific Generative Adversarial Networks (GANs) to generate missing views.

Main Results:

• GIMPP effectively handles missing views in multi-view datasets.
• The model demonstrated superior performance over state-of-the-art methods.
• Experiments were conducted on TCGA-BRCA and METABRIC breast cancer datasets.

Conclusions:

• The developed GIMPP model successfully addresses the missing view problem in multi-view prediction.
• Explicitly generating missing data enhances breast cancer prognosis prediction.
• The approach shows significant promise for bioinformatics and healthcare applications.