Abstract
The selective expression of genes is the basis of cellular and individual phenotypic differences, serving as a mediator in the causal pathways from genotypes to phenotypes. Single-cell differential expression analysis identifies distinct transcriptomic landscapes, but fails to establish causal relationships due to the presence of confounders. On the other hand, causal inference methods in population genetics such as Mendelian randomization often overlook the heterogeneity among cells and dynamic changes along trajectory. To address these limitations, we propose the trajectory-inference-based dynamic single-cell Mendelian randomization (ti-scMR), integrating population genomes and single-cell transcriptomes to explore transcriptional features causally linked to cellular and individual phenotypes. ti-scMR leverages trajectory inference and functional principal component analysis to capture the temporal cumulative effects of gene expression, select genetic instrumental variables through single-cell expression quantitative trait locus (eQTL) mapping, and employ transcriptome-level Mendelian randomization to prioritize causal genes for cellular and individual phenotypes, specifically those that are related through affecting cellular development. We demonstrate the superiority of ti-scMR in identifying causal genes through simulations. With application in two real single-cell datasets, we discover potential causal genes on immune cell differentiation and related disease. The integration of single-cell trajectory inference, eQTL, and Mendelian randomization will make ti-scMR a powerful tool for elucidating the causal mechanisms underlying complex traits.