Systematic Deciphering of ATBC Nephrotoxicity Mechanisms via Machine Learning and Single‐Cell Analysis

ABSTRACT Persistent environmental pollutants are an emerging concern for human health. Acetyl tributyl citrate (ATBC) is a widely used plasticizer with potential renal toxicity, yet the molecular basis of ATBC‐related kidney injury remains incompletely understood. Here, we present an integrative computational framework with ATBC‐specific target prioritization and multilayer validation to generate testable hypotheses. First, we curated 201 reported ATBC‐interacting targets and identified 131 targets detectable in a public bulk transcriptomic dataset of acute kidney injury (AKI). Using a target‐prioritized differential analysis, we selected 15 ATBC targets (primary set) and 5 highly significant nontarget genes as controls (secondary set), yielding 20 genes for downstream modeling and validation. We then trained multiple machine‐learning classifiers and implemented an enhanced validation suite, including repeated cross‐validation, 1000‐iteration bootstrap resampling to derive performance confidence intervals, SHAP‐based model interpretability, calibration analysis, decision‐curve analysis, and feature‐stability assessment. Finally, we validated cell‐type‐specific expression patterns of the 20 genes using single‐cell RNA sequencing data from GSE199321 (human AKI and ∼60,000 cells) with appropriate statistical testing. Collectively, our analyses suggest that ATBC exposure may be associated with dysregulation of pathways related to renal vascular function, tubular injury responses, and xenobiotic transport. Importantly, all findings are computational and should be interpreted as hypothesis‐generating; experimental studies are required to establish causality and to validate the predicted targets and mechanisms.