conformeR: conformalized differential expression analysis of multi-condition single-cell data

Differential expression (DE) analysis in multi-condition single-cell transcriptomics poses several challenges: extensive multiple testing across tens of thousands of genes and cells, the need for interpretable inference at the cell-type and population level, and robustness to model assumptions. conformeR tackles these issues by combining conformal inference with counterfactual prediction to generate valid p-values for DE testing. Rather than performing gene-wise analyses, conformeR exploits conditional dependencies between genes to predict counterfactual expression levels, from which conformal prediction intervals for treatment effects are constructed. These intervals are then used to derive p-values that quantify evidence against the null hypothesis of no differential expression. P-values are subsequently transformed into local false discovery rates, then into their frequentist counterparts, and aggregated across cells and patients to achieve rigorous and interpretable FDR control at the cell-type level. At its current stage, conformeR shows encouraging results on both toy and real datasets. Ongoing work focuses on improving computational efficiency, strengthening integration with existing counterfactual prediction tools, and scaling to larger datasets. This work was first presented at EuroBioC 2025 in Barcelona.