Inference of Genetic Networks from Pseudo Time Series of Single-cell Gene Expression Data using Modified Random Forests.

This study proposes a novel method for inferring genetic networks using both steady-state and pseudo time-series data of single-cell gene expressions. While several methods for inferring genetic networks from time series of bulk-cell gene expression data have been proposed, many of these approaches use time derivatives of gene expression levels. However, since pseudo time-series data lack precise temporal information about when measurements were taken, time derivatives cannot be calculated from this data. Therefore, existing methods are ineffective for analyzing pseudo time-series data. To address this limitation, our proposed method does not use time derivatives of gene expression levels but uses their signs. We theorize that, even when no precise temporal information is available, the signs of time derivatives, which indicate whether the gene expression levels are increasing or decreasing, can be estimated from pseudo time-series data. Our approach was designed on the basis of GENIE3 and its extensions, which, although essentially intended to infer genetic networks from bulk-cell gene expression data, have reportedly performed well in this respect. Validation through numerical experiments with both artificial and real gene expression data confirms the effectiveness of our proposed method.