Effects of pseudobulk and Gaussian noise on the application of the dynamical network biomarker theory to single-cell RNA-seq data

Abstract

The dynamical network biomarker (DNB) theory detects early warning signals at the transition state (pre-disease state) in complex biological systems. The DNB theory has been applied to microarray data in several diseases. However, this theory has not yet been extensively to limited single-cell RNA sequencing (scRNA-seq) data. The main problem arises from missing values causing a standard deviation of zero, resulting in the calculation of correlation coefficients impossible in the DNB theory. The present study introduces pseudobulk and Gaussian noise to missing values in the scRNA-seq data to avoid division by zero. Without compromising the data characteristics, these two techniques detected previously missed genes for the DNB analysis and, thus, successfully expanded the scope of the DNB theory.