Machine Learning-Based Screening of Cosmetic Ingredients Identifies Vat Blue 6 as a Thyroid Hormone Receptor β Disruptor

Thyroid disorders are among the most prevalent endocrine
conditions
worldwide, exhibiting a rising incidence and disproportionately affecting
women. In this study, we hypothesized that cosmetics may contain previously
unidentified thyroid-disrupting chemicals. To evaluate this possibility,
we compiled a comprehensive data set of cosmetic ingredients and developed
a random forest regression-based machine learning model to predict
their potential to disrupt thyroid hormone receptor β (TRβ),
a critical regulator of thyroid function. From the top 40 compounds
ranked by the model, 12 frequently used cosmetic ingredients were
selected for experimental validation. Of these, six demonstrated measurable
binding affinity toward TRβ. Notably, Vat Blue 6 (VB6), a colorant
utilized in cosmetic formulations, exhibited structural characteristics
potentially mimicking thyroid hormones and displayed potent TRβ
binding with an affinity (Kd) as low as
0.7 μM. Subsequent in vitro assays and in vivo experiments in mice confirmed VB6’s thyroid-disrupting
effects, evidenced by dose-dependent reductions in serum thyroid hormone
concentrations and morphological alterations of thyroid tissue. This
study highlights the efficacy of machine learning approaches in rapidly
screening large chemical inventories to identify potential thyroid
disruptors and underscores the critical need for further toxicological
assessment of cosmetic ingredients, particularly considering their
frequent and prolonged exposure among female populations.