Environmental and Molecular Mutagenesis Health, Toxicology and Mutagenesis

Defining an Objective Inconclusive Zone to Enhance PCA ‐Based Classification in the TGx ‐ DDI and TGx ‐ HDACi Biomarkers

Andrew Williams, Julie K. Buick, Jiri Aubrecht, Florence Burleson, Christine E. Crute, Yasmin Dietz‐Baum, Lena Dorsheimer, Albert J. Fornace, Roland Frötschl, Constance A. Mitchell, Xiaotong Wang, Lorreta Yun‐Tien Lin, Heng‐Hong Li, Carole L. Yauk
Wiley · 2026-10-01 · DOI 10.1002/em.70078

Abstract

ABSTRACT TGx‐DDI and TGx‐HDACi are transcriptomic biomarkers used to classify chemicals based on gene expression profiles indicative of DNA damage‐inducing (DDI) or histone deacetylase inhibition (HDACi) mechanisms. Both employ nearest shrunken centroid (NSC) analysis, hierarchical cluster analysis (HCA), and principal component analysis (PCA). While NSC allows for inconclusive calls when class membership probability is low, PCA has historically lacked an objectively defined inconclusive region. This study establishes a distance‐based probabilistic threshold for PCA, using TGx‐DDI as the primary example, to define a quantitative inconclusive zone along the first principal component (PC1). Samples with ambiguous PC1 values are objectively classified as inconclusive, preventing overconfident misclassifications of weak transcriptomic signals. For TGx‐HDACi, the inconclusive region is narrower but widens substantially with fewer biomarker genes, highlighting increased uncertainty with limited transcriptomic data. Implementing these thresholds enhances the reliability of TGx‐DDI assessments, transparently aligning PCA with NSC to inform regulatory decision‐making.

Read at publisher →