Integration of incomplete multi-omics data using Knowledge Distillation and Supervised Variational Autoencoders for...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Cancer Research

Background:

• High-throughput technologies generate diverse omics data (mRNA, DNA methylation, microRNA) for disease study.
• Integrating multi-omics data enhances understanding of cancer's molecular basis and disease progression prediction.
• Conventional methods struggle with high-dimensional omics data and the curse of dimensionality.

Purpose of the Study:

• To develop a novel framework for effective multi-omics data integration and cancer progression prediction.
• To address the challenges of high dimensionality and limited common samples in multi-omics datasets.
• To improve the accuracy of predicting patient survival outcomes in various cancer types.

Main Methods:

• Introduced Knowledge Distillation and Supervised Variational AutoEncoders utilizing View Correlation Discovery Network (KD-SVAE-VCDN).
• Applied the KD-SVAE-VCDN framework to integrate high-dimensional multi-omics data.
• Evaluated the model's performance on breast and kidney carcinoma datasets for survival prediction.

Main Results:

• The KD-SVAE-VCDN architecture accurately predicted disease progression in breast and kidney carcinoma.
• The model effectively classified patients into long-term and short-term survivor groups.
• KD-SVAE-VCDN demonstrated superior performance compared to state-of-the-art multi-omics integration models.

Conclusions:

• The KD-SVAE-VCDN framework shows efficacy in predicting cancer progression and patient survival outcomes.
• This approach supports personalized medicine by enabling tailored treatment strategies.
• The model's performance suggests potential for advancing cancer research and clinical management.