Gene Selection in a Single Cell Gene Space Based on D-S Evidence Theory

Area of Science:

• Computational Biology
• Bioinformatics
• Data Science

Background:

• High-dimensional single-cell gene expression data presents challenges due to noise and redundancy, leading to uncertainty.
• Traditional methods like Bayesian probability theory have limitations in handling such uncertain data.
• Dempster-Shafer (D-S) evidence theory offers a robust framework for managing uncertainty with weaker conditions than Bayesian approaches.

Purpose of the Study:

• To develop and evaluate a gene selection strategy for single-cell gene spaces to mitigate noise and redundancy.
• To leverage D-S evidence theory for quantifying and managing uncertainty in gene expression data.
• To improve classification and clustering performance in single-cell analysis through effective gene selection.

Main Methods:

• Defining cell-to-cell distance based on individual gene expression values.
• Establishing tolerance relations derived from the defined cell distances.
• Applying belief and plausibility functions, based on tolerance classes, to model single-cell gene space uncertainty.
• Developing and implementing gene selection algorithms utilizing these belief and plausibility functions.

Main Results:

• The developed belief and plausibility functions effectively measure the uncertainty inherent in single-cell gene spaces.
• The proposed gene selection algorithms demonstrate superior performance compared to existing state-of-the-art methods.
• The algorithm achieves significant improvements in classification and clustering accuracy.
• A high gene reduction rate was observed, indicating efficient noise and redundancy removal.

Conclusions:

• D-S evidence theory provides a powerful tool for addressing uncertainty in high-dimensional single-cell gene expression data.
• The proposed gene selection method effectively reduces noise and redundancy, enhancing downstream analysis.
• This approach offers a promising advancement for improving the accuracy and efficiency of single-cell data analysis, particularly for classification and clustering tasks.