Abstract
Investigating how genes jointly affect complex human diseases is important, yet challenging. The network approach (e.g., weighted gene co-expression network analysis (WGCNA)) is a powerful tool. However, genomic data usually contain substantial batch effects, which could mask true genomic signals. Paired design is a powerful tool that can reduce batch effects. However, it is currently unclear how to appropriately apply WGCNA to genomic data from paired design. In this paper, we modified the current WGCNA pipeline to analyse high-throughput genomic data from paired design. We illustrated the modified WGCNA pipeline by analysing the miRNA dataset provided by Shiah et al. (2014), which contains forty oral squamous cell carcinoma (OSCC) specimens and their matched non-tumourous epithelial counterparts. OSCC is the sixth most common cancer worldwide. The modified WGCNA pipeline identified two sets of novel miRNAs associated with OSCC, in addition to the existing miRNAs reported by Shiah et al. (2014).
Introduction
In genomic research, "batch effects"—noise introduced by different lab conditions, reagent lots, or individual patient variability—can easily drown out real biological signals. A "paired design" is a powerful experimental setup where researchers collect two samples from the same individual (like a tumor and the adjacent healthy tissue). This naturally controls for the patient's unique genetic background and environment.
However, the gold-standard tool for finding gene networks, WGCNA, was designed with the assumption that every sample is independent. When you have paired data, the samples are mathematically "correlated" because they come from the same person. Applying standard WGCNA to paired data without modification can lead to statistical inaccuracies or missed discoveries. This paper provides a formal framework to bridge that gap by integrating Linear Mixed-Effects Models (LMM) into the workflow.
Methodology
The authors proposed a five-step modified WGCNA pipeline. The most significant technical shifts include:
- Validation of Pearson Correlation: The authors mathematically demonstrated that the Pearson correlation coefficient remains a valid measure for co-expression even when data is paired.
- Linear Mixed-Effects Models (LMM): This is the core innovation. In standard WGCNA, researchers relate gene "modules" (clusters) to a disease trait using simple correlation. In the modified version, the authors use LMM to account for the "within-pair" correlation. By treating the subject as a random effect, the model can more accurately determine if a gene module is truly associated with the disease status (cancer vs. normal) across the whole population.
- Gene Significance (GS) Calculation: Instead of simple correlation, GS is calculated as the absolute value of the test statistic from the LMM, providing a more robust measure of a gene's importance to the phenotype in a paired context.
Results and Discussion
The study validated its approach using a publicly available miRNA dataset (GSE45238) consisting of 40 specimens of Oral Squamous Cell Carcinoma (OSCC) and 40 matched non-tumorous epithelial counterparts. The analysis identified four distinct miRNA modules: turquoise (254 miRNAs), blue (189), brown (78), and grey (309).
As shown in the module-trait relationship heatmap, the turquoise module showed the strongest association with cancer status ($t\text{-value} = -18.68, p < 1.97 \times 10^{-20}$). Interestingly, the grey module, which usually contains unassigned genes, was also highly significant ($t\text{-value} = 10, p < 4 \times 10^{-12}$). This suggests that many OSCC-related miRNAs are not necessarily co-expressed with each other but are still individually critical.
Further validation of the turquoise module revealed a high correlation (cor=0.53, p=8.9e-24) between module membership and gene significance for disease status, indicating that hub genes in this module are likely key drivers of the disease phenotype. Specifically, the miRNA miR-let-7c was identified as the hub of the turquoise module, connected to 140 other miRNAs. This is biologically significant as the let-7 family is well-known as a tumor suppressor.
Conclusion
The modified WGCNA pipeline identified 71 of the 84 miRNAs previously reported in the original study, while also uncovering two entirely new sets of miRNAs. While the sample size of 40 pairs is respectable, further investigation in independent cohorts is necessary to confirm these novel associations. Future work could explore whether these highly connected "hub genes" can be used as reliable clinical biomarkers to predict the risk or stage of oral cancer.
Summary updated