2dFDR for confounder adjustment

Saakshi Bangera, DY Patil School of Biotechnology and Bioinformatics

Genomic profiling technologies of a high-throughput level generate a massive number of omics data. This is because such technologies enable biological system analysis at different omics levels. Statistical analysis of omics data tests the relationship between omics aspects and a covariate of interest. Observational omics studies are exposed to numerous types of confounding. 

Confounding occurs when the association between the primary variable and the omics feature is distorted by a confounder (another variable) due to its relationship with both.

Frequent confounders of omics studies are- 

1. Demographic variables such as age, race, and gender. 
2. Biological heterogeneity such as a cell mixture in a tissue sample
3. Technical variation (batch effect) occurs when samples are not processed together.

By introducing an extra variable, confounding reduces the statistical power. It also increases the probability of false findings. 

Standard approaches to tackle confounding include stratification and regression. Adjusting the confounders reduces the statistical power. It is usual when no significant relationships are recovered after adjusting confounders. Thus, the statistical power should be improved despite being under confounding and multiple testing. This topic has the potential to rescue an underpowered study. Confounders may affect only a certain part of omics features. Adjusting it for every feature will lead to an extended power loss. A power-friendly method to adjust a confounder is to test its significance first. If it is not significant the confounder is excluded from the regression model. This strategy improves power but fails to control the type 1 error properly. 

This study takes a different approach to this problem and integrates the confounder adjustment into multiple testing frameworks (FDR control). In this approach, the unadjusted statistic screens out the features that are less probable to be linked to the confounder or the covariate of interest. Next, FDR control is performed based on the adjusted statistic. Two-dimensional false discovery rate control procedure (2dFDR) has proved to offer asymptomatic FDR control. 

Overview of the 2dFDR

2dFDR is constructed on linear models and is a common modelling approach for operational omics data. The outcome of this approach is the measurement of the omics feature and it assumes that the confounders are known. It depends on the unadjusted and adjusted test statistics and proceeds in two dimensions. In the first dimension, 2dFDR uses the statistic for the adjusted analysis to screen out the irrelevant omics features (ones that are not associated with the confounder). In the second dimension, true signals on the remaining features are identified using an adjusted statistic. This also helps to control the FDR at a required level. The unadjusted statistic can be regulated to increase the signal density and reduce multiple testing loads in the second dimension. When the correlation between the variable of interest and the confounder is high, the signals and noises overlap on the adjusted test statistic. This is due to loss of power with confounder adjustment. For achieving the desired FDR level, 2dFDR uses the unadjusted statistic to exclude the irrelevant features first. After that, a much lower adjusted statistic cutoff is used to achieve the exact same FDR level. This allows the model to achieve significant power improvement which increases with the correlation between the confounder and variable of interest. 

Significance of the study

The 2dFDR procedure is quite powerful when the confounder and the variable of interest are highly correlated. 2dFDR also works excellently when the confounder affects only a subset of an omics feature. In this study, confounder adjustment is integrated into multiple testing. 2dFDR offers asymptomatic FDR control and enhances the power of traditional procedures based on the adjusted statistic. 2dFDR is considerably more powerful than conventional procedures. It is a robust tool and has broad potential in omics association studies. 

Also read: A novel approach to reduce graft-vs-host disease

Source: Yi, S., Zhang, X., Yang, L. et al. 2dFDR: a new approach to confounder adjustment substantially increases detection power in omics association studies. Genome Biol 22, 208 (2021). https://doi.org/10.1186/s13059-021-02418-8 

• The Corrosion Prediction from the Corrosion Product Performance
• Nitrogen Resilience in Waterlogged Soybean plants
• Cell Senescence in Type II Diabetes: Therapeutic Potential
• Transgene-Free Canker-Resistant Citrus sinensis with Cas12/RNP
• AI Literacy in Early Childhood Education: Challenges and Opportunities

About the author: Saakshi is currently pursuing MSc in Biotechnology from DY Patil School of Biotechnology and Bioinformatics. She believes that she doesn’t have a specific area of interest yet. She wishes to explore toxicology and food biotechnology. She’s quite passionate about Biotechnology and aims to grab every opportunity she comes across.