Julia Faillace Thiesen , Elise Jacquemet , Pascal Campagne, Denis Chatelain, Etienne Brochot, Yves‑Edouard Herpe , Nolwenn M. Dheilly , Fabrice Bouilloux, Bénédicte Rognon, Alexandre Douablin, Guillaume Leboucher , Florent Percher, Marc Eloit and Philippe Pérot
- Faillace Thiesen et al. Molecular Medicine
- https://doi.org/10.1186/s10020-025-01238-x
Abstract
Background
Cervical cancer screening programs are increasingly relying on sensitive molecular approaches as primary tests to detect high‑risk human papillomaviruses (hrHPV), the causative agents of cervix cancer. Although hrHPV infection is a pre‑requisite for the development of most precancerous lesions, the mere detection of viral nucleic acids, also present in transient infections, is not specific of the underlying cellular state, resulting in poor positive predictive values (PPV) regarding lesional states. There is a need to increase the specificity of molecular tests for better stratifying individuals at risk of cancer and to adapt follow‑up strategies.
Methods
HPV‑RNA‑SEQ, a targeted RNA next generation sequencing assay allowing the detection of up to 16 hrHPV splice events and key human transcripts, has previously shown encouraging PPV for the detection of precancerous lesions. Herein, on 302 patients with normal cytology (NILM, n = 118), low‑grade (LSIL, n = 104) or high‑grade squamous intraepithelial lesions (HSIL, n = 80), machine learning‑based model improvement was applied to reach 2‑classes (NILM vs HSIL) or 3‑classes (NILM, LSIL, HSIL) predictive models.
Results
Linear (elastic net) and nonlinear (random forest) approaches resulted in five 2‑class models that detect HSIL vs NILM in a validation set with specificity up to 0.87, well within the range of PPV of other competing RNA‑based tests in a screening population.
Conclusions
HPV‑RNA‑SEQ improves the detection of HSIL lesions and has the potential to complete and eventually replace current molecular approaches as a first‑line test. Further performance evaluation remains to be done on larger and prospective cohorts.