Nonlinear ridge regression improves cell-type-specific differential expression analysis

Abstract Background Epigenome-wide association studies (EWAS) and differential gene expression analyses are generally performed on tissue samples, which consist of multiple cell types. Cell-type-specific effects a trait, such as disease, the omics interest but difficult or costly to measure experimentally. By measuring data for bulk tissue, type composition sample can be inferred statistically. Subsequently, cell-type-specific estimated by linear regression that includes terms representing interaction between proportions trait. This approach involves two issues, scaling multicollinearity. Results First, although is analyzed in scale, methylation/expression suitably logit/log scale. To simultaneously analyze scales, we applied nonlinear regression. Second, show highly collinear, obstructive ordinary cope with multicollinearity, ridge regularization. In simulated data, attained well-balanced sensitivity, specificity precision. Marginal model lowest precision highest sensitivity was only algorithm detect weak signal real data. Conclusion Nonlinear test performance. The omicwas package R implements EWAS, QTL analyses. software freely available from https://github.com/fumi-github/omicwas