Phenotype Classification Using Moment Features of Single-Cell Data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Standard gene expression classification relies on averaged cell data.
• Single-cell analysis offers individual cell expression measurements, enabling richer feature vectors.
• Higher-order moments and mixed moments can be derived from single-cell data.

Purpose of the Study:

• To demonstrate the utility of expression moments from single-cell data for cancer classification.
• To compare the performance of different moment-based features in classification tasks.
• To evaluate the impact of feature selection and sample size on classification accuracy.

Main Methods:

• Utilized synthetic data from mammalian cell cycle and melanoma pathway networks.
• Employed real single-cell data from HT-29 and HCT-116 cell lines.
• Generated data from hidden binary regulatory networks with Gaussian observations.
• Calculated and applied mean, variance, skewness, and mixed moments for classification.
• Performed feature selection and analyzed sample size effects.

Main Results:

• Classification accuracy improved progressively with the inclusion of variance, skewness, and mixed moments over just the mean.
• Analysis of real single-cell data for EGFR and ERRB2 genes showed improved classification using higher-order moments (mean, variance, skewness).
• Feature selection and sample size influenced classification performance.

Conclusions:

• Expression moments derived from single-cell data provide superior features for cancer classification compared to traditional methods.
• The inclusion of higher-order and mixed moments significantly enhances classification performance.
• This methodology offers a promising avenue for advancing precision oncology and biomarker discovery.