Genetic data visualization using literature text-based neural networks: Examples associated with myocardial...

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Effective data visualization is crucial for complex, high-dimensional datasets, particularly in biology and medicine.
• Existing visualization methods struggle with large genetic datasets, often limited by dimensionality and sensitivity to missing data.
• Interpretable visualization of genetic information, like single nucleotide polymorphisms (SNPs), is essential but challenging.

Purpose of the Study:

• To develop a novel literature-based visualization method for high-dimensional genetic data.
• To preserve both global and local structures of single nucleotide polymorphisms (SNPs) during dimensionality reduction.
• To enable interpretable visualizations by integrating textual information from scientific literature.

Main Methods:

• Proposed a literature-based approach to reduce high-dimensional SNP data.
• Utilized literature text representations to preserve SNP dynamics and enable textual interpretability.
• Evaluated the method using machine learning models for classifying risk factors (race, myocardial infarction, sex).
• Employed visualization techniques for data clustering and quantitative metrics for performance evaluation.

Main Results:

• The proposed method successfully reduced data dimensionality while preserving global and local SNP structures.
• Achieved interpretable visualizations by leveraging textual information from literature.
• Outperformed popular dimensionality reduction and visualization methods in classification and visualization tasks.
• Demonstrated robustness against missing and higher-dimensional genetic data.
• Showcased feasibility in integrating genetic data with other risk information from literature.

Conclusions:

• The literature-based visualization method offers a powerful tool for analyzing large genetic datasets.
• It enhances interpretability and classification accuracy, outperforming existing techniques.
• The method is robust to data imperfections and adaptable for integrating diverse risk information.