Interpretable Raman phenotyping of human granulosa cells aligns with endocrine-defined ovarian status.

Assessment of ovarian status in in vitro fertilization (IVF) is critically reliant on systemic endocrine and morphological markers, which often lack the temporal resolution and mechanistic specificity to predict individual oocyte competence, necessitating a proximal, label-free molecular phenotype. Here, we present a high-definition, label-free Raman microspectroscopy pipeline for the non-destructive phenotyping of human granulosa cells (GCs)-the essential somatic niche of the oocyte. By profiling 140 participants using a participant-anchored analytical framework and machine learning, we identified a robust biochemical axis that discriminates between good ovarian status (GOS) and poor ovarian status (POS) cohorts. GOS cells are characterized by significantly higher intensities in lipid (1444, 2856, and 2928 cm-1) and amide bands, reflecting enhanced steroidogenesis and membrane remodeling. Conversely, POS cells exhibit a relative prominence in nucleic-acid-linked features (915 and 1180 cm-1), suggesting altered biosynthetic balance or stress signatures. Our convolutional neural network models leveraged this molecular gradient to achieve a participant-level Receiver-Operating Characteristic (ROC) area of 0.882 on a blinded test cohort. These spectral markers showed concordance with established endocrine indices (AMH and FSH), anchoring the Raman phenotype to endocrine-defined ovarian status. This study therefore establishes interpretable granulosa-cell Raman phenotyping as a reproducible molecular readout aligned with endocrine stratification. Further validation against clinically meaningful reproductive outcomes and in external multicenter cohorts will be required to determine translational value beyond current endocrine assessment.