Interpretable machine learning prioritizes transcriptomic features associated with ferroptosis-mitochondrial permeability transition net activity states in cervical cancer.

CombinedNet transcriptomic scoring method shows moderate-to-high concordance with existing pathway activity measures and recovers promoter-inhibitor balance in cervical cancer data.

• Why it matters: Understanding the joint features of ferroptosis and mitochondrial permeability transition in cervical cancer can clarify stress-related transcriptional signals involved in carcinogenesis, but their pathway-state features remain poorly characterized.
• What they did: Researchers developed a directionality-aware net pathway activity framework using gene sets related to ferroptosis and MPT, evaluated it with simulations and external datasets, and applied machine learning with feature prioritization and robustness assessments.
• The result: CombinedNet achieved an AUC of 0.780 with significant permutation test p-value, and prioritized CKMT1B and KLK13 as key features, but found no stable association with lesion stage or FIGO stage, suggesting the score's limited utility as a lesion-stage biomarker.