Abstract 5095: Statistical modeling of transcriptional regulatory states in single-cell RNA-Seq data of tumor and infiltrated immune cells

• Published in: Cancer research (Chicago, Ill.) Vol. 79; no. 13_Supplement; p. 5095
• Main Authors: Wan, Changlin, Chang, Wennan, Lu, Xiaoyu, Sun, Yifan, So, Kaman, Cao, Sha, Lu, Xiongbin, Zhang, Chi
• Format: Journal Article
• Language: English
• Published: 01.07.2019
• DOI: 10.1158/1538-7445.AM2019-5095
• ISSN: 0008-5472

Single-cell RNA-seq of tumor tissue has greatly strengthened our understanding of tumor development as well as the heterogeneous tissue micro-environment. Many models are dedicated to detect deferentially expressed genes among single cells of different conditions, or to discover (novel) sub-cell types via various clustering methods. However, current statistical models often neglect the zero/low expressions, and use one single modality to fit the rest of the non-zero ones. We believe that zero and low expressions are observed partly due to limited experimental resolution; on the other hand, occasionally, low expressions are falsely observed as non-zero due to the existence of incompletely degraded mRNA. In addition, the severe heterogeneity and fast dynamics of single cells makes it invalid to fit their expressions in one single modality. Here, we propose a left truncated mixture Gaussian (LTMG) distribution to uncover the multimodality in single cells’ gene expression while properly handling the zero and low expressions. LTMG is motivated by the kinetic relationships among the mRNA abundance, transcriptional regulatory signals (TRSs), and mRNA metabolism in a cell. Specifically, observations of a gene’s expression are fitted as mixture Gaussian distributions which assumes zero and low expressions as left censored data. By treating each mixing component as one state of the transcriptional regulators, we combined our LTMG model and our in-house bi-clustering algorithm to infer modules of genes co-regulated by certain (unknown) transcriptional regulatory signal, which is only shared by a subset of all the single cells. Application of the method to three high quality single-cell RNA-seq data sets of T cells extracted from liver, colon and lung cancer micro-environments, identified varied transcriptional regulations specific to subclasses of T cells. Citation Format: Changlin Wan, Wennan Chang, Xiaoyu Lu, Yifan Sun, Kaman So, Sha Cao, Xiongbin Lu, Chi Zhang. Statistical modeling of transcriptional regulatory states in single-cell RNA-Seq data of tumor and infiltrated immune cells [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 5095.