Micronucleus quantification from whole-slide haematology images using AI serves as a translatable pharmacodynamic biomarker for DNA damage response inhibitors.

Micronuclei are widely recognised biomarkers of genomic instability and DNA damage, making their accurate quantification essential for understanding the pharmacodynamic properties of chemotherapeutic agents and inhibitors of the DNA damage response (DDR). Here, we report the development and validation of a novel assay for the automated detection and quantification of micronuclei within circulating red blood cells (RBC) from peripheral blood smears. We integrate recent advances in whole-slide imaging (WSI) technologies and supervised deep-learning algorithms to quantify micronuclei in over 100,000 RBCs from a single image. We demonstrate that this approach achieves strong analytical concordance with flow cytometry (Pearson’s r = 0.926, P < 0.0001) while offering distinct advantages. Additionally, using May-Grünwald Giemsa dyes we show that deep-learning algorithms can stratify red blood cells into both mature erythrocytes and immature reticulocytes from WSIs. Critically, we establish that micronuclei-positive red blood cell (MN+-RBC) frequency correlates with anti-tumor efficacy in BRCA1-deficient xenograft models following exposure to PARP inhibitors and demonstrates dose-dependent pharmacodynamic (PD) responses. Furthermore, we show that whole-slide imaging offers several advantages over widely used flow cytometry approaches, including the identification of cells with multiple micronuclei and the ability to quantify morphological features associated with detrimental pre-analytical conditions. These findings position automated WSI-based micronucleus quantification as a scalable, minimally invasive PD biomarker requiring only 5 μl of blood that enables longitudinal monitoring of DDR inhibitor therapies.