Fig 2: Classification of cervical swab samples’ cytology categories based on the beta-actin-normalized VCP and p16INK4a concentrations. (A,C) Predicted versus ground truth classifications mapped to biomarker-to-beta-actin ratios. (B,D) The corresponding ROC curves. (A,B) demonstrate the results for the classifier fitted to the training data including the normal, RCC, LSIL, and HSIL cases. (C,D) correspond to the results of the same model fitted to the same training data including the ASC-US cases. The classification was performed using a multinomial GAM. Prediction regions (A,C) and ROC curves (B,D) were based on bootstrap cross-validation with 200 replications, using 25% of cases in each cytology class as the test set for each replication. The areas under the curve (AUCs) in (B) were 79.9% (normal), 83.2% (RCC), 91.8% (LSIL), and 97.9% (HSIL). For (D), the AUCs were 73.1% (normal), 87.5% (RCC), 71.4% (ASC-US), 72.4% (LSIL), and 95.2% (HSIL).

Fig 3: Classification of cervical tissues’ histopathological diagnoses based on the proportion of MCM2- and VCP-positive nuclei. (A) Predicted versus ground truth classifications mapped to the proportion of tumor cells expressing MCM2 and VCP. The training data included LSIL, HSIL, and SCC cases. (B) The corresponding ROC curves (areas under the curve: LSIL = 98.9%, HSIL = 87.7%, and SCC = 95.6%). Prediction regions (A) and ROC curves (B) were based on bootstrap cross-validation with 200 replications, using 33.3% of the data in each diagnosis class as the test set for each replication.

Fig 4: Biomarker expression reported as ratios to beta-actin in lysates of CC cell lines and PCS cells. (A) ELISA measurements of TOP2A, MCM2, VCP, and p16INK4a normalized to beta-actin in lysates of HeLa, Ca Ski, HT-3, and C-33 A cancer cell lines, and PCS cells. Statistical significance was assessed using a Gamma-distributed log-link GLMM, with fixed effects for proteins, cell types, and their interactions, and random effects for biological replicates (n = 3). (Holm-corrected pairwise t-test: *: p < 0.05, **: p < 0.01, ***: p < 0.001, ****: p < 0.0001. Boxplots represent the median (bar), interquartile range (IQR, box), and 1.5 × IQR (whiskers); the points correspond to individual data points.) (B) PCA of biomarker concentrations across cultured CC cell lines and PCS cells.

Fig 5: Immunostaining and quantification of biomarker expression in tissues of cervical SILs and CC subtypes. Quantification of (A) TOP2A, (B) MCM2, (C) VCP, and (D) p16INK4a in IHC tissue stainings of cervical SILs (LSIL; n = 9, and HSIL; n = 12) and carcinomas (GCC; n = 2–3, and SCC; n = 29–32). The quantitative analysis was performed by measuring the proportion of positive nuclei for each biomarker in TMAs of different histopathology diagnoses. Each point represents an individual case, with point size being proportional to the number of nuclei in the core and logit transformation applied to proportions for improved visualizations. Statistical significance was assessed using a logit-link quasibinomial generalized linear model. (Holm-corrected pairwise t-test: *: p < 0.05, **: p < 0.01, ****: p < 0.0001. Boxplots represent the median (bar), interquartile range (IQR, box), and 1.5 × IQR (whiskers).) (E) Representative images of original IHC-stained sections (scale bars = 200 μm) and their computationally pseudocolor-coded images (scale bars = 100 μm) that were quantified using the Visiopharm software, version 2024.07.2.17285x64. Brown represents the original IHC staining, purple denotes the nuclear counterstain, while pseudocolors indicate staining intensity: red (strong), orange (moderate), and yellow (weak).