The results of Transcriptome-wide Mendelian Randomization (TWMR) in large-scale populations can directly validate, across scales, the results of causal inference from deep learning combined with double machine learning on single-cell transcriptomes of human samples.