The Coherence Response Index: predicted microRNA targets show no coherent regulation in chemically exposed mouse embryonic stem cells.

Abstract

Toxicogenomic studies increasingly promote microRNA (miRNA) expression changes as chemical-toxicity biomarkers and infer their meaning from computationally predicted messenger RNA (mRNA) targets, rarely testing those predictions against matched mRNA data. We present the Coherence Response Index (CRI), an open, expression-matched check of whether a responsive miRNA's predicted targets shift in the expected direction. CRI recovered target repression in two positive controls: miRNA-mimic transfection (6/6 productive; AUC≈0.73) and, at endogenous stoichiometry in mouse embryonic stem cells (mESCs), Dicer1 knockout (137 of 412 evaluable expressed miRNAs productive, against 35 of 211 non-expressed controls); spiked-in coherent shifts of 0.1 log2 units were detected in 77% of tests. We then applied CRI to two of our own published datasets from the same RNA preparations of mESCs exposed for 24 h to nine chemicals (paired small- and long-RNA-seq, duplicate). Across 225 miRNA×chemical tests, predicted targets showed no coherent regulation (AUC mean 0.489, SD 0.050); a single borderline call (miR-200a-5p, DEHP) was not robust to resampling. Restricting the background to protein-coding genes added significant shifts, but not in the miRNA's direction (8 coherent, 12 anti-coherent). Separately, the published fold changes of the mapped benzene biomarkers do not reproduce (sign agreement 54%, Spearman ρ = 0.11): they arise from isoform-level quantification and a zero-replacement log-ratio estimator that, on identical data, yields thousands of extreme values that vanish under a pseudocount estimator. In this short-exposure mESC screening design, predicted-target enrichment did not describe observed regulation; CRI offers a reusable check wherever a matched mRNA layer exists.