Overcoming biases in causal inference of molecular interactions

Exact model-free function inference using uniform marginal counts for null population

MOTIVATION

Recognizing cause-effect relationships is a fundamental inquiry in science. However, current causal inference methods often focus on directionality but not statistical significance. A ramification is chance patterns of uneven marginal distributions achieving a perfect directionality score.

RESULTS

To overcome such issues, we design the uniform exact function test with continuity correction (UEFTC) to detect functional dependency between two discrete random variables. The null hypothesis is two variables being statistically independent. Unique from related tests whose null populations use observed marginals, we define the null population by an embedded uniform square. We also present a fast algorithm to accomplish the test. On datasets with ground truth, the UEFTC exhibits accurate directionality, low biases, and robust statistical behavior over alternatives. We found nonmonotonic response by gene TCB2 to beta-estradiol dosage in engineered yeast strains. In the human duodenum with environmental enteric dysfunction, we discovered pathology-dependent anti-co-methylated CpG sites in the vicinity of genes POU2AF1 and LSP1; such activity represents orchestrated methylation and demethylation along the same gene, unreported previously. The UEFTC has much improved effectiveness in exact model-free function inference for data-driven knowledge discovery.

AVAILABILITY AND IMPLEMENTATION

An open-source R package "UniExactFunTest" implementing the presented uniform exact function tests is available via CRAN at doi: 10.32614/CRAN.package.UniExactFunTest. Computer code to reproduce figures can be found in supplementary file "UEFTC-main.zip."

Integrated analysis of bulk and single-cell RNA-seq reveals the role of MYC signaling in lung adenocarcinoma

MYC is one of the well-known oncogenes, and its important role in cancer still remains largely unknown. We obtained lung adenocarcinoma (LUAD) multi-omics data including genome, transcriptome, and single-cell sequencing data from multiple cohorts. We calculated the GSVA score of the MYC target v1 using the ssGSEA method, and obtained the genes highly correlated with this score by Spearman correlation analysis. Subsequent hierarchical clustering divided these genes into two gene sets highly associated with MYC signaling (S1 and S2). Unsupervised clustering based on these genes divided the LUAD samples into two distinct subgroups, namely, the MYC signaling inhibition group (C1) and activation group (C2). The MCP counter package in R was used to assess tumor immune cell infiltration abundance and ssGSEA was used to calculate gene set scores. The scRNA-seq was used to verify the association of MYC signaling to cell differentiation. We observed significant differences in prognosis, clinical characteristics, immune microenvironment, and genomic alterations between MYC signaling inhibition and MYC signaling activation groups. MYC-signaling is associated with genomic instability and can mediate the immunosuppressive microenvironment and promote cell proliferation, tumor stemness. Moreover, MYC-signaling activation is also subject to complex post-transcriptional regulation and is highly associated with cell differentiation. In conclusion, MYC signaling is closely related to the genomic instability, genetic alteration and regulation, the immune microenvironment landscape, cell differentiation, and disease survival in LUAD. The findings of this study provide a valuable reference to revealing the mechanism of cancer-promoting action of MYC in LUAD.