Precision Oncofertility: Integrating Genomics, Reproductive Biomarkers, Treatment Toxicity Profiles, and Foundation Models for Fertility Preservation in Men and Women with Cancer
Year: 2026
Type: article
Study design: Observational Study
Abstract: Background: Cancer survivors of reproductive age face substantial risks of treatment-induced gonadotoxicity, yet clinical decision-making for fertility preservation remains largely empirical and population-based.A precision oncofertility framework that integrates genomic, biomarker, and toxicity data is urgently needed.Objective: To develop and validate a multimodal foundation model that integrates genomic variants, sex-specific reproductive biomarkers, treatment toxicity profiles, and clinical variables to predict fertility impairment risk and support individualized fertility-preservation decisions in cancer patients.Methods: A retrospective observational cohort of 400 patients (age 18-45 years) diagnosed with breast cancer, lymphoma, leukemia, testicular cancer, ovarian cancer, cervical cancer, or colorectal cancer was assembled.All patients received fertility-preservation counselling prior to treatment.Data included reproductive biomarkers (anti-Müllerian hormone (AMH), antral follicle count (AFC), follicle-stimulating hormone (FSH), estradiol; sperm concentration, motility, testosterone, inhibin-B), genomic features (BRCA1, BRCA2, CHEK2, TP53, FMR1), treatment toxicity variables (alkylating agent and platinum exposure, radiation dose, stem cell transplantation), and standard clinical covariates.A transformer-based foundation model with four specialised encoders and cross-modal attention was developed.Performance was benchmarked against Random Forest, XGBoost, and a Deep Neural Network using accuracy, precision, recall, F1 score, and area under the receiver-operating-characteristic curve (AUC).Results: The foundation model achieved the highest performance across all metrics: accuracy 0.883, precision 0.876, recall 0.862, F1 0.869, and AUC 0.934.AMH (importance score 0.187), AFC (0.171), and alkylating agent exposure (0.152) were the strongest predictors.Fertility risk stratification identified 37.0% (n=148) low-risk, 35.5% (n=142) moderate-risk, and 27.5% (n=110) high-risk patients.BRCA1/2 mutations and radiation dose independently amplified impairment risk.Conclusion: A precision oncofertility framework incorporating multimodal data and transformer-based architecture substantially outperforms conventional machine-learning approaches.AI-guided risk stratification can enable individualized, evidence-based fertility-preservation counselling and intervention, with direct implications for clinical oncofertility practice.
Language: English
Subfield: Reproductive Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W7221507060