Data-driven detection of subtype-specific and differentially expressed genes

Abstract

Tissue or cell subtype-specific and differentially-expressed genes (SDEGs) are defined as being differentially expressed in a particular tissue or cell subtype among multiple subtypes. Detecting SDEGs plays a critical rolse in molecularly characterizing and identifying tissue or cell subtypes, and facilitating supervised deconvolution of complex tissues. Unfortunately, classic differential analysis assumes a convenient null hypothesis and associated test statistic that is subtype-non-specific and thus, resulting in a high false positive rate and/or lower detection power with respect to particular subtypes. Here we introduce One-Versus-Everyone Fold Change (OVE-FC) test for detecting SDEGs. To assess the statistical significance of such test, we also propose the scaled test statistic OVE-sFC together with a mixture null distribution model and a tailored permutation scheme. Validated with realistic synthetic data sets on both type 1 error and detection power, OVE-FC/sFC test applied to two benchmark gene expression data sets detects many known and de novo SDEGs. Subsequent supervised deconvolution results, obtained using the SDEGs detected by OVE-FC/sFC test, showed superior performance in deconvolution accuracy when compared with popular peer methods.

COT: an efficient Python tool for detecting marker genes among many subtypes

“…In relation to previous work, the effort to detect SMG can be traced back to One-Versus-Rest (OVR) test (Yu, Feng et al 2010), One-Versus-One (OVO) test (Hunt, Freytag et al 2019), and most recently One-Versus-Everyone (OVE) test (Wang, Hoffman et al 2016). We and others have recognized that the test statistics used by most existing methods do not exactly satisfy SMG definition (1) and often require ad hoc OVE set intersection (Yu, Feng et al 2010, Hunt, Freytag et al 2019, Chen, Lu et al 2020.…”

“…More importantly, COT framework is efficient in that neither OVE set intersection nor intractable sample permutation is needed in estimating the null distribution (Hunt, Freytag et al 2019, Chen, Lu et al 2020, where a significant majority of genes are assumed to be associated with the null hypothesis. The null distribution plays a crucial role in large-scale multiple testing.…”

“…The null distribution plays a crucial role in large-scale multiple testing. However, because the size of subtype-enriched samples is often relatively small and the non-SMG patterns are highly complex, classical methods to estimate the null distribution in a two-sample test setting is impractical or intractable (Chen, Lu et al 2020) and…”

“…and Multiple Comparisons with the Best (MCB) (Hsu 1996, Wang, Master et al 2006, Newman, Liu et al 2015. Importantly, we and others have recognized that the test statistics used by most existing methods do not exactly satisfy SMG definition and often ad hoc OVE set intersections have been used to finalize SMG (Wang, Hoffman et al 2016, Chen, Lu et al 2020, Patrick, Taga et al 2020.…”

“…We implement and test the latest functionalities of COT workflow in Python through a three-step assessment. We have previously demonstrated the successful applications of COT prototype on gene expression and proteomics data (Yu, Feng et al 2010, Chen, Lu et al 2020.…”