A quantitative prediction method utilizing whole omics data for biosensing.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Omics data offer rich information for biomarker discovery.
• Current prediction methods struggle with omics data's high dimensionality and noise.
• Underutilization of omics data limits biomarker development.

Purpose of the Study:

• To develop a robust quantitative prediction method for biomarker construction using omics data.
• To address the challenges of data multidimensionality and noise in omics analysis.
• To enhance the utility of omics data for identifying physiological and ecological biomarkers.

Main Methods:

• Developed OmicSense, a quantitative prediction method utilizing a mixture of Gaussian distributions.
• Benchmarked OmicSense using a transcriptome dataset.
• Employed weighted gene co-expression network analysis to assess interpretability.
• Applied OmicSense to single-cell transcriptome, metabolome, and microbiome datasets.

Main Results:

• OmicSense demonstrated accurate and robust prediction against background noise without overfitting in transcriptome data.
• Weighted gene co-expression network analysis showed OmicSense utilizes hub nodes, indicating interpretability.
• High prediction performance (r > 0.8) was achieved across single-cell transcriptome, metabolome, and microbiome datasets.
• The method proved applicable to diverse omics data types and scientific fields.

Conclusions:

• OmicSense is an effective quantitative prediction tool for biomarker discovery from omics data.
• The method overcomes limitations of existing approaches, offering robustness against noise and interpretability.
• OmicSense facilitates the accelerated use of omics data as biosensors across various scientific domains.