Multi-insight visualization of multi-omics data via ensemble dimension reduction and tensor factorization

Area of Science:

• Data Visualization
• Bioinformatics
• Computational Biology

Background:

• High-dimensional data visualization is crucial for exploratory data analysis and knowledge discovery.
• Current dimensionality reduction (DR) techniques offer limited perspectives, risking the omission of important data features.
• Relying on a single DR projection can be misleading due to method-specific strategies and diverse low-dimensional representations.

Purpose of the Study:

• To propose the first framework for multi-insight data visualization specifically designed for multi-omics data.
• To overcome the limitations of single-insight approaches by uncovering a majority of data features.
• To provide a more comprehensive understanding of complex biological datasets.

Main Methods:

• Developed a framework that combines multiple dimensionality reduction (DR) methods.
• Utilized tensor factorization to integrate diverse DR techniques.
• Employed clustering to group solutions into an optimal number of insights.

Main Results:

• The proposed framework effectively uncovers a majority of data features through multiple insights.
• Experimental evaluations demonstrated a competitive advantage over state-of-the-art methods.
• Validated on synthetic, simulated ovarian cancer, and real breast cancer multi-omics data.

Conclusions:

• The multi-insight visualization framework offers a more comprehensive approach to analyzing multi-omics data.
• This method enhances the discovery of biological insights compared to traditional single-projection techniques.
• The framework shows significant potential for advancing multi-omics data exploration in cancer research.